Prediction of insufficient serum vitamin D status in older women: a validated model

Prediction of insufficient serum vitamin D status in older women: a validated model Summary We developed an externally validated simple prediction model to predict serum 25(OH)D levels < 30, < 40, < 50 and 60 nmol/L in older women with risk factors for fractures. The benefit of the model reduces when a higher 25(OH)D threshold is chosen. Introduction Vitamin D deficiency is associated with increased fracture risk in older persons. General supplementation of all older women with vitamin D could cause medicalization and costs. We developed a clinical model to identify insufficient serum 25-hydroxyvitamin D (25(OH)D) status in older women at risk for fractures. Methods In a sample of 2689 women ≥ 65 years selected from general practices, with at least one risk factor for fractures, a questionnaire was administered and serum 25(OH)D was measured. Multivariable logistic regression models with backward selection were developed to select predictors for insufficient serum 25(OH)D status, using separate thresholds 30, 40, 50 and 60 nmol/L. Internal and external model validations were performed. Results Predictors in the models were as follows: age, BMI, vitamin D supplementation, multivitamin supplementation, calcium supplementation, daily use of margarine, fatty fish ≥ 2×/week, ≥ 1 hours/day outdoors in summer, season of blood sampling, the use of a walking aid and smoking. The AUC was 0.77 for the model using a 30 nmol/L threshold and decreased in the models with higher thresholds to 0.72 for 60 nmol/L. We demonstrate that the model can help to distinguish patients with or without insufficient serum 25(OH)D levels at thresholds of 30 and 40 nmol/L, but not when a threshold of 50 nmol/L is demanded. Conclusions This externally validated model can predict the presence of vitamin D insufficiency in women at risk for fractures. The potential clinical benefit of this tool is highly dependent of the chosen 25(OH)D threshold and decreases when a higher threshold is used. . . . . . Keywords Aged Decision support techniques Logistic models Osteoporosis Vitamin D Vitamin D deficiency * T. Merlijn Introduction tmerlijn@gmail.com Vitamin D supplements are widely used, either self- Department of General Practice and Elderly Care Medicine, administered or with a prescription [1]. The widespread use Amsterdam Public Health Research Institute, VU University Medical is not surprising as vitamin D supplementation for groups at Center, Amsterdam, The Netherlands risk for vitamin D insufficiency is advised in most internation- Stichting ArtsenLaboratorium en Trombosedienst, Koog aan de al and national guidelines. These guidelines find their base in Zaan, The Netherlands the potential harmful effect of low vitamin D status [2–4]. Department of Internal Medicine and Endocrinology, Amsterdam Traditionally, vitamin D deficiency is associated with the dis- Public Health Research Institute, VU University Medical Center, Amsterdam, The Netherlands eases rickets and osteomalacia. Nowadays, these conditions are rare in daily practice. Many vitamin D-related association Department of Health Sciences, Faculty of Earth and Life Sciences, Amsterdam Public Health Research Institute, VU University studies and intervention trials have been conducted in the last Amsterdam, Amsterdam, the Netherlands decades. The observational studies have shown that vitamin D Department of Epidemiology and Biostatistics, Amsterdam Public status is inversely correlated with multiple diseases and con- Health Research Institute, VU University Medical Center, ditions such as fractures, falls, cardiovascular disease, Amsterdam, The Netherlands 1540 Osteoporos Int (2018) 29:1539–1547 different forms of cancer and psychiatric disorders. However, Methods intervention trials have so far only shown positive effects of vitamin D supplementation on bone-related problems and not For the development of our model, we followed the instruc- on other outcomes [5]. Supplementation of vitamin D in com- tion of the TRIPOD [26]. The steps of our method are pre- bination with calcium can reduce loss of bone mineral density sented schematically in Fig. 1. and reduces fracture risk [6–10]. Reduction of fractures by supplementation of both calcium and vitamin D is particularly Study population clear in the oldest group frail persons [11]. Vitamin D with or without calcium supplementation might also reduce falls, but Participants were selected from a randomised prospective trial meta-analyses have shown contradictory results [12–17]. regarding the detection and treatment of older women at risk The serum 25-hydroxyvitamin D (25(OH)D) concentration for fractures in general population and subsequent treatment is the best marker of vitamin D status. The threshold for the with anti-osteoporotic medication, the SALT Osteoporosis need of vitamin D supplementation is still under debate. While Study (SOS) [21]. In this study, women with ages between some advice treatment at 25(OH)D levels of < 25–40 nmol/L, 65 and 90 years were enrolled from GP-practices surrounding others claim that this should be < 75 nmol/L [18, 19]. The five laboratory locations in Noord-Holland, a province in Health Council of the Netherlands considers a level > The Netherlands (Northern Latitude 52°). Women with at least 50 nmol/L sufficient for women older than 50 years and one risk factor for fractures (fracture after the age of 50 years, men older than 70 years and advises daily vitamin D supple- parental hip fracture, low body weight, immobility and con- mentation for this entire group [4]. The Institute of Medicine ditions that may cause secondary osteoporosis) who had been (IOM) has proposed the same threshold. The IOM presumes randomised to the intervention group of the study were eval- that the average requirement of the population is covered with uated with a questionnaire, bone densitometry and laboratory serum 25(OH)D levels of 40 nmol/L [2]. examination [27]. Informed consent was obtained from all When using 50 nmol/L as cut-off, about 50% of the Dutch individual participants and the study was approved by the older persons has a vitamin D insufficiency [20, 21]. Dutch Health Council (OGZ 2.978.265). Considering that only half of all older persons have serum For the development of the prediction model, we used the 25(OH)D levels below the recommended threshold, a general baseline data from all women in the intervention group collected advice to supplement vitamin D in these elderly is only useful in the period from March 2010 until February 2013. An addi- for half of the older population. In the other half of the popu- tional vitamin D-related questionnaire was filled out by the par- lation, treatment would not be necessary, depending on the ticipants that were included during this period. The women who season of blood collection. When lower thresholds are used, had their examination in Amsterdam were excluded because in the number of overtreatment would be even more obvious. this laboratory, a different 25-hydroxyvitamin D assay was used. The consequences are medicalization and costs [22]. Furthermore, we excluded Black, Arabic, Turkish and Asian Testing of serum 25(OH)D levels can reduce the number of women (n = 103) and women in a residential care home (n = unnecessary treatments but is also expensive and the outcome is 24), resulting in a total of 2689 participants. The reason for ex- influenced by the period of the year. Three prediction models to cluding these women was that there were too few participants in predict vitamin D deficiency in elderly were published before these groups and most of them already had an indication for [23–25]. In an analogous manner, we wanted to develop a pre- vitamin D supplementation. diction model to predict insufficient vitamin D status in older For the external validation of the prediction model, we collect- women. In contrast to previous studies, our study focuses on ed data from a second population in the same region. All women the clinical implementation of the model. Therefore, we focused between 65 and 90 years old who had been referred by a general on older women with risk factors for fractures and evaluated practitioner for combined dual X-ray absorptiometry (DXA) and clinical applicability. In the second place, we used a spectrum laboratory examination from September 2010 until February of thresholds, in a way that medical professionals can choose the 2013 were asked to participate and to complete the same ques- threshold they consider to be appropriate. tionnaires (n = 856). This was a non-overlapping population in The goal of this study was to develop a validated prediction the same region that was evaluated in the same laboratory. model that could discriminate older women who do or do not need treatment with vitamin D supplements. We aimed to Questionnaires construct a simple tool that can be added to fracture risk eval- uation in general practice or that can be used as a self-test. All women completed two questionnaires. The first question- While there is still discussion about the optimal cuff of value naire contained questions about risk factors for fractures and of 25(OH)D levels, we will develop the model with different osteoporosis. If a questionnaire was returned incomplete, the thresholds for serum 25(OH)D levels, namely 30, 40, 50 and participant was contacted by telephone to complete the miss- 60 nmol/L. ing answers. The second questionnaire contained questions Osteoporos Int (2018) 29:1539–1547 1541 Fig. 1 Scheme of the Populaon women ≤65 years development and validation of the prediction model ≥1 risk factor for fractures Quesonnaire, biometry, and serum 25(OH)D Exclusion: -Women in nursing home - Black, Arabic and Asian women Data exploraon/ preparaon: -Choosing cut-offs in connuous variables -Opmalizaon modeling of the seasons -Imputaon if more than 5% missings Predicon models: logisc regression with backward selecon (p<0.157) Four thresholds: 30, 40, 50, and 60 nmol/L Internal validaon of the 4 models Simplificaon of regression coefficients to 4 risks scores External validaon primary models and risk scores Calculaon of predicve values of models and examples of praccal use about predictors of vitamin D status. The questionnaires were immobility, use of walking aid, falling during previous collected at the laboratory visit when DXA and blood tests 12 months, vitamin D supplementation either self- were performed. administered or prescribed, multivitamin supplementation, calcium supplementation, smoking, consumption of fatty fish, Laboratory analyses consumption of margarine, alcohol use, time spending outside in winter and summer, the level of education and Serum 25(OH)D was analysed with a chemoluminescence polypharmacy. assay (Diasorin, Stillwater, MN, USA) as a routine measure- Age was calculated as date of examination minus date of ment at the Centrum voor Medische Analyse Antwerpen. The birth. Weight and height were measured during the visit of the laboratory is accredited and meets the European norm ISO laboratory and BMI was calculated as kg/m . 15180. The inter-assay coeficient of variation was 10.1% at All other questions were self-reported. Patients were con- a 25(OH)D concentration of 61.5 nmol/L and 9.9% at a sidered to be immobile when severe problems with walking 25(OH)D concentrations of 36 nmol/L. were indicated (yes/no). Use of walking aid, current smoking, falls (at least 1 in the last 12 months), supplementation of Outcome vitamin D, calcium and multivitamin use (daily) were dichot- omous questions (yes/no). In a second question, the partici- In the models, we used a dichtome outcome: below and equal pants were asked to specify whether the supplementation of to or above the threshold level of 25(OH)D of, respectively, vitamin D was on prescription or self-administered. 30, 40, 50 and 60 nmol/L. Time spent outdoors in winter and summer, level of edu- cation, alcohol use, fatty fish and margarine use were mea- Predictors of vitamin D status sured in four categories. For the model, these categories were dichotomized. The selected potential predictors were age and body mass Medication use was collected with the question to list all actual index (BMI), period of the year of blood sampling, medication. We defined polypharmacy as the use of six drugs or 1542 Osteoporos Int (2018) 29:1539–1547 more. We used individual dates of blood samples for serum For the external validation, we calculated the AUC of 25(OH) measurement to determinate the period of the year. models and the practical model based on the risk score in a second population. Model development Examples of application The development of the prediction model was preceded by the We shall show two examples how the prediction model can be preparation of some potential predictors. Before modelling, we used for a practical purpose. In the first place, we will produce evaluated collinearity of the variables. There was none. We used figures with predictive values for the different thresholds. That splines to evaluate non-linear relationships between continuous might be useful for predicting the probability for deficiency in predictors and the outcome and to modify continuous variables daily practice. In the second place, we apply this tool to esti- into categories. In order to reduce the variables for the period of mate the proportion of women at risk for 25(OH) insufficiency the year, we observed the pattern of seasonal change of 25(OH)D that need supplementation in a specific period of the year. levels over the months during 3 years. We used univariate linear regression of 25(OH)D levels to find an optimal reflection of the Software seasonal influence in a maximum of four periods and selected the model with the highest R and the least amount of periods. IBM SPSS Statistics version 22 was used for data checking To select the predictors for the model, we performed logistic and modelling and R 3.2.2 was used for bootstrapping and regression analyses. We used a backward selection strategy. That assessment of the final models. means that we started with all potential predictors in the model, and every turn, the predictor with the highest P value was ex- cluded from the model, until all P values were lower than our Results selection criterion of p<0.157 (Akaike’s information criterion). A stricter p value leads to the development of models that are Population characteristics closely adapted to the data and that generalise poorly. Since missing data can effect a prognostic model, we Between March 2010 and February 2013, we collected ques- planned to use multiple imputation to estimate the missing tionnaires and serum 25(OH)D values of 2689 women. Of values according to the Multivariate Imputation by Chained these women, 2624 (97.6%) had complete data; hence, impu- Equation procedure in IBM SPSS Statistics 20, only in case of tation was not necessary. Serum 25(OH)D levels ≥ 60, ≥ 50, ≥ more than 5% missing cases. Due to few missing cases, im- 40 and ≥ 30 nmol/L were found in 32, 50, 65 and 90% of the putation was not necessary. women, respectively. The population for external validation comprised 856 women. The main characteristics of both pop- Model performance ulations are shown in Table 1. To evaluate the discrimination of the model, i.e., if the model Modelling of seasonal changes is able to distinguish patients with and without an insufficient 25(OH)D status, a receiver operator characteristic (ROC) Figure 2 shows the seasonal change of the mean serum curve was made and the area under the roc curve (AUC) 25(OH)D levels. The model with the highest R and the lowest calculated. The goodness-of-fit of the model was tested by number of periods was a model with three variables: the nadir the Hosmer-Lemeshow test. The unexplained variance was in December till April, a peak in July and August and inter- indicated by Nagelkerke’s R . mediate levels in the rest of the months. Internal validation was performed with bootstrapping tech- niques. Regression coefficients and performance of the Development of the prediction models models were adjusted according to the optimism estimates from the internal validation procedure. Irrespective of the chosen threshold, the best prediction model We constructed a risk score by multiplying the regression contained the following predictors: age, BMI, walking aid, coefficients of the predictors by 10 and divided by 3. The vitamin D supplementation either self-administered or pre- result was rounded to the nearest whole number. The last step scribed, multivitamin use, calcium supplementation, smoking, was made to keep the scores low to keep the model practical. time spent outdoors in summer and period of blood sample. At To check the loss of information, we compared AUC before thresholds of 30 and 40 nmol/L, use of margarine and fatty and after dividing by 3. We calculated the sensitivity, speci- fish were predictors. From a threshold of 50 nmol/L, use of ficity, positive predictive value and negative predicted value fatty fish was not a predictor and use of margarine was a very for the defined thresholds. weak predictor and disappeared from the model at higher Osteoporos Int (2018) 29:1539–1547 1543 Table 1 The prevalence of the determinant in the study population, the regression coefficients of the models with different cut-offs and corresponding risk scores Model 1: threshold Model 2: threshold Model 3: threshold Model 4: threshold 30 nmol/L 40 nmol/L 50 nmol/L 60 nmol/L Primary population External sample Regression Risk Regression Risk Regression Risk Regression Risk N = 2689 N = 856 coefficient (B) score coefficient (B) score coefficient (B) score coefficient (B) score 1 2 3 4 25-hydroxyvitamin D (nmol/L) 51.8 (20.5) 54.7(20.9) < 30 14.1% 11.3% < 40 30.7% 27.6% < 50 50.4% 54.3% < 60 68.1% 60.4% Age (years) 73.5(6.1) 73.1 (6.0) 65–70 32.6% 34.2% 70–75 27.3% 29.0% 0.30 1 0.32 1 0.34 1 > 75 40.0% 36.8% 0.59 2 0.75 2 0.69 2 0.67 2 Body mass index (kg/m ) 28.0 (6.1) 27.1 (4.5) < 25 28.7% 33.5% 25–30 40.9% 41.0% 0.27* 1 0.22 1 0.27 1 > 30 30.4% 25.5% 0.44 1 0.61 2 0.52 2 0.44 1 Impaired mobility (% yes) 26.0% 17.4% 0.24* 1 Walking aid (% yes) 26.2% 18.3% 0.63 2 0.58 2 0.48 2 0.45 1 No of falls in last year (% ≥ 1) 29.9% 46.0% 0.20 1 0.15* 1 Vitamin D supplementation (% 17.3% 30.4% yes) Self-administered 11.0% 15.4% − 0.98 − 3 − 0.72 − 2 − 0.74 − 2 − 0.41 − 1 Prescribed 6.3% 15.0% − 0.88 − 3 − 1.36 − 5 − 1.55 − 5 − 1.16 − 4 Use of multivitamins (% yes) 24.4% 23.4% − 0.52 − 2 − 0.79 − 3 − 0.92 − 3 − 0.65 − 2 Calcium supplementation (% yes) 19.7% 29.1% − 0.38 − 1 − 0.44 − 1 − 0.52 − 2 − 0.45 − 1 Fattyfishconsumption(% 12.7% 12.1% − 0.43 − 1 − 0.27 − 1 ≥ 2×/week) Margarine consumption daily (% 69.6% 69.2% − 0.54 − 2 − 0.32 − 1 − 0.15* − 1 yes) Smoking (% yes) 10.2% 11.9% 0.52 2 0.30* 1 0.30 1 0.21* 1 Use of alcohol (% ≥ 1 unit/day) 25.7% 27.5% Time outdoors in winter (% 22.9% 21.7% − 0.23* − 1 − 0.24 − 1 > 60 min/day) Time outdoors in summer (% 80.3% 77.6% − 0.67 − 2 − 0.63 − 2 − 0.50 − 2 − 0.36 − 1 > 60 min/day) Month of examination (%) Dec-Apr 34.0% 45.1% 1.62 5 1.50 5 1.26 4 1.19 4 May-Jun or Sep-Nov 42.8% 42.1% 0.83 3 0.79 3 0.70 2 0.63 2 Jul-Aug 23.3% 12.8% Fracture > 50 year of age (% yes) 45.6% 55.1% Parental hip fracture (% yes) 26.9% 13.6% No of medication (% > 6) 17.5% 15.4% Education (% none or low) 80.0% 79.5% Constant − 2.53 − 1.52 − 0.56 0.18 *Significance level of p < 0.154 All other regression coefficients significance level of p < 0.05 1544 Osteoporos Int (2018) 29:1539–1547 sensitivity, specificity, positive predictive value and neg- ative predictive value of the model. The risk scores for the different models are shown in Table 1. Application of the risk score To use the model, one has to choose a threshold. Figure 3 shows the predictive value for the different thresholds. For example, when a threshold of 50 nmol/ L is desired and a patient has a risk score of 0 or lower, this patient has a 70% probability to have 25(OH)D levels above the threshold. Table 3 shows the proportion women at risk for 25 (OH)D insufficiency in the primary population without vitamin D supplementation. The proportion is presented for the different thresholds, a desired probability and per season. For example, to have a probability of 80% of having 25(OH)D levels above 50 nmol/L, 100% are at Fig. 2 Mean serum 25(OH)D levels per month. Error bars show 95% CI risk in winter, 94% in spring and autumn and 79% in of the mean summer. This shows that the model could only differen- tiate few women who do not need treatment at this thresholds. Falling on the other hand was a predictor at the threshold. However, at a threshold of 40 nmol/L, the higher thresholds but not at 30 nmol/L. Time spent outdoor in model selects 88% at risk in winter and 50% in spring winter was a predictor at thresholds of 40 and 50 nmol/L. and autumn. What means that the model is able to se- Table 1 shows the β’s of the different predictors at the differ- lect 50% of the population that would not need treat- ent thresholds. ment most of the year. It is clear that at a threshold of 30 nmol/L, even more women not at risk could be se- Validation of the prediction model lected but nihil at a threshold of 60 nmol/L. The internally validated models showed an AUC of 0.72 to 0.77 with the highest for the threshold 30 nmol/L. There is little loss of information when the model is Discussion converted to a risk score. The R is 0.25 for the thresh- old of 40 nmol/L. The AUC of the external validation In this study, we present four validated models for the predic- was between 0.71 and 0.82 (Table 2). Figure 3 shows the tion of vitamin D status for different 25(OH)D thresholds Table 2 AUCofthe ROCcurve AUC after Nagelkerke AUC in and Nagelkerke R square of the internal R square external internal validated models and the validation sample simplified models for the different 25(OH)D thresholds Model threshold 30 nmol/L Prediction model 0.77 0.21 0.82 Risk score 0.77 0.21 0.82 Model threshold 40 nmol/L Prediction model 0.76 0.25 0.75 Risk score 0.76 0.25 0.74 Model threshold 50 nmol/L Prediction model 0.75 0.24 0.72 Risk score 0.75 0.24 0.72 Model threshold 60 nmol/L Prediction model 0.73 0.18 0.71 Risk score 0.72 0.18 0.71 AUC area under the curve, ROC receiver operator characteristics For the simplified models, the regression coefficients were multiplied by 3/10 and rounded to the nearest whole number Osteoporos Int (2018) 29:1539–1547 1545 Model 1: 25(OH)D ≤ 30nmol/L Model 2: 25(OH)D ≤ 40nmol/L 100% 100% 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 Model 4: 25(OH)D ≤ 60nmol/L Model 3: 25(OH)D ≤ 50nmol/L 100% 100% 90% 90% 80% 80% PPV 70% 70% NPV 60% 60% 50% 50% Specificity 40% 40% 30% 30% Sensivity 20% 20% Prevalance 10% 10% 0% 0% -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Fig. 3 Four models with different thresholds of serum 25-hydroxyvitamin D. The Y-axis shows the positive predictive (PPV), negative predictive value (NPV), sensitivity, specificity and prevalence for any computed risk score (X-axis) of participants in primary population between 30 and 60 nmol/L. In our primary sample and in an 25(OH)D was tested. Some predictors were not in all models. external sample, reasonable predictive values were shown. The intake of fatty fish is only present in the models with The evaluation of the models with different thresholds eluci- threshold 30 and 40 nmol/L, respectively. The use of marga- dates that the model with a threshold of 60 nmol/L has less rine is a predictor in the models with thresholds from 30 to predictors and lower prediction than the other models. 50 nmol/L, but is a weak predictor at 50 nmol/L. This suggest Furthermore, we have shown that the models with thresholds that food-related vitamin D intake helps to prevent the severest of 50 and 60 nmol/L have low negative predictive value and deficiency, but not to maintain higher values. therefore are less potent to exclude low 25(OH)D levels in our Most predictors are similar to those observed in earlier population, due to the high prevalence of 25(OH)D levels studies. The AUC in other studies for a threshold of lower than these thresholds. The highest AUC were seen at 50 nmol/L range from 0.73 to 0.86 [17–19]. Only one study thresholds of 30 and 40 nmol/L. (AUC of 0.71) was validated in an external population [17]. Predictors for insufficient 25(OH)D in all models were The age of the participants in these studies was similar, but higher age, higher BMI, use of walking aid, little time spent there was difference in the prevalence of 25(OH)D deficiency. outdoors in summer, smoking, no calcium supplementation, In contrast to our model, other studies added more complex no use of multivitamins, no use of vitamin D on prescription predictors that consist of more than one question or a physical or self-administered and the period of the year when serum examination. Table 3 Application of the models: percentage of women at risk for 25(OH)D insufficiency per season for different thresholds of 25(OH)D and probability to reach the threshold Threshold 25(OH)D Model 1 30 nmol/L Model 2 40 nmol/L Model 3 50 nmol/L Model 4 60 nmol/L Percentage above threshold 70% 80% 90% 70% 80% 90% 70% 80% 90% 70% 80% 90% 25(OH)D insufficiency Not at all 100% 84% 20% 35% 12% 1% 3% 0% 0% 0% 0% 0% Winter 0% 16% 80% 65% 88% 99% 97% 100% 100% 100% 100% 100% Spring and autumn 0% 0% 18% 4% 50% 89% 79% 94% 100% 99% 100% 100% Summer 0% 0% 0% 0% 0% 51% 27% 79% 100% 99% 99% 99% Calculated in the participants without vitamin D supplementation in the primary population 1546 Osteoporos Int (2018) 29:1539–1547 We have observed that the use of calcium supplements is simple measurement that lacks in our model that might improve associated with higher 25(OH)D levels, which can be explained the performance of the model is physical activity [23, 24]. by a lower vitamin D metabolite turnover due to reduction of the Information on physical activity was not available in our study, parathyroid hormone levels [28]. but there may be an overlap with impaired mobility, the use of a With respect to vitamin D supplementation, it is remarkable walking aid and the time that is spent outdoors [23]. Time spent that vitamin D on prescription is a stronger predictor than self- outdoors in summer as well in winter are positive predictors. administered vitamin D. There are two possible explanations. Because of the low intensity of the sun in winter, we do not think First, the compliance might be better when vitamin D is given that this is a direct effect, but that it might be a reflection of on prescription. Secondly, it is possible that patients obtain physical activity and general health as well [31]. lower dosages without prescription. The 25 (OH)D assay we used was not compared with the The period of the year is one of the strongest predictors in our reference of the national institute of standards and technology. model. As our data covers 3 years, we observed that there were Nevertheless, we used a single 25(OH)D assay with good perfor- substantial differences per year. In most studies, the change in mance [30]. serum 25(OH)D levels is supposed to be a symmetric sinusoidal In conclusion, the proposed model can help to distinguish curve over the seasons. However, in our data, the mean levels women with adequate serum 25(OH)D levels in a population of over 3 years show a pattern in which the rise of serum 25(OH)D older women with risk factors for fractures. This might reduce in spring is steeper than the decrease in autumn, a narrow peak in unnecessary treatment with vitamin D in some relatively healthy summer and a wide dip in winter. The result of this pattern is that older women. To reach levels above the 50 nmol/L during the the optimal model does not exactly follow the seasons. The peak whole year, practically all women need supplementation and is in the months of July and August; the dip is from December many of them need continuous supplementation. If one allows until the end of April. The other months are between these sum- levels of 30 or 40 nmol/L, the model can substantially reduce the mer and winter levels. A similar pattern has been reported earlier number of women that need supplementation, at least during a in a Danish cohort of postmenopausal women [29]. part of the year. This shows that advices in guidelines can have A strength of this study is that it was performed in a large substantial impact on the number of women with an indication for population of women at risk for fractures and that it was val- supplementation. idated in an external population of women with comparable Acknowledgements The SALT Osteoporosis Study has been largely age. In the external validation, the loss of discrimination was funded by Stichting Achmea Gezondheidszorg. Healthcare costs have limited or absent, indicating that the model is robust. been compensated by Achmea and VGZ Zorgverzekeraar. Additional There are different opinions about the optimal threshold for financial support has been provided by Stichting ArtsenLaboratorium serum 25(OH)D [18, 19]. Also, one might differ about wheth- en Trombosedienst. The sponsors do not have any role in the design or implementation of the study, data collection, data management, future er one prefers a high specificity with a high positive predictive data analysis and interpretation, or in the preparation, review or approval value, but more false negatives or one prefers a high sensitiv- of the manuscript. ity with a high negative predictive value but more false posi- tives. Therefore, we have presented our data in a way that Compliance with Ethical Standards medical professionals can choose between different thresholds and between different positive and negative predictive values. Informed consent was obtained from all individual participants and the study The selection of a population of women with increased was approved by the Dutch Health Council (OGZ 2.978.265). fracture risk has advantages and disadvantages. A profit is that Conflicts of interest The authors declare that they have no conflict of we have exact insight in the performance of the model in a interest. population in which 25(OH)D levels are clinically most im- portant. A disadvantage is that the application of the model Ethical approval All procedures performed in studies involving human has only been validated in Dutch Caucasian women with in- participants were in accordance with the ethical standards of the institu- creased fracture risk, and therefore, we must be careful with a tional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. broader implementation of this model. Unfortunately, we did not have enough data of non-Caucasian women living in the Open Access This article is distributed under the terms of the Creative Netherlands to gain more insight in these specific groups. Commons Attribution-NonCommercial 4.0 International License (http:// creativecommons.org/licenses/by-nc/4.0/), which permits any noncom- Another limitation is the lack of information on the dose of mercial use, distribution, and reproduction in any medium, provided vitamin D supplementation. Furthermore, a large part of variance you give appropriate credit to the original author(s) and the source, pro- is not explained. The unexplained variance might be reduced by vide a link to the Creative Commons license, and indicate if changes were improvement of the accuracy of the measurements, e.g., with the made. use of standardised intake questionnaires or the measurement of ultraviolet exposure. However, for our purpose, a simple tool for daily practice, more complex measurements are not suitable. One Osteoporos Int (2018) 29:1539–1547 1547 16. Bolland MJ, Grey A, Reid IR (2014) Differences in overlapping References meta-analysis of vitamin D supplements and falls. J Clin Endocinol Metab 99(11):4265–4272. https://doi.org/10.1210/jc.2014-2562 1. Kantor ED, Rehm CD, DU M, White E, Giovannucci EL (2016) 17. Murad MH, Elamin KB, Elnour OA, Elamin MB, Alkatib AA, Trends in dietary supplement use among US adults form 1999- Fatourechi MM et al (2011) The effect of vitamin D on falls: a 2012. JAMA 316(14):1464–1474. https://doi.org/10.1001/jama. systematic review and meta-analysis. J Clin Endocrinol Metab 2016.14403 96(10):2997–3006. https://doi.org/10.1210/jc.2011-1193 2. Ross AC, Manson JE, Shapses SA (2011) The 2011 report on 18. Reid IR (2014) Skeletal and nonskeletal effects of vitamin D: is dietary reference intakes for calcium and vitamin D from the insti- vitamin D a tonic for bone and other tissues? Osteoporos Int tute of medicine: what clinicians need to know. J Clin Endocrinol 25(10):2347–2357. https://doi.org/10.1007/s00198-014-2749-7 Metab 96(1):53–58. https://doi.org/10.1210/jc.2010-2704 19. Bischoff-Ferrari HA (2010) Vitamin D and fracture prevention. 3. Wood CL, Cheetham TD (2016) Vitamin D: increasing use among Endocrinol Metab Clin N Am 39(2):347–353. https://doi.org/10. at-risk groups (NICE guideline PH56). Arch Dis Child Educ Ed 1016/j.ecl.2010.02.009 101(1):43–45. https://doi.org/10.1136/archdischild-2015-308299 20. Van Schoor NM, Visser M, Pluijm SM, Kuchuk N, Smit JH, Lips P 4. Health Council of the Netherlands (2012) Evaluation of the dietary (2008) Vitamin D deficiency as a risk factor for osteoporotic frac- reference values for vitamin D. Health Council of the Netherlands, tures. Bone 42(2):260–266. https://doi.org/10.1016/j.bone.2007.11. The Hague 5. Theodoratou E, Tzoulaki I, Zgaga L, Ioannidis JP (2014) Vitamin D 21. Janssen HCJP, Emmelot-Vonk MH, Verhaar HJJ, van der Schouw and multiple health outcomes: umbrella review of systematic re- YT (2013) Determinants of vitamin D status in healthy men and views and meta-analyses of observational studies and randomised women aged 40-80 years. Maturitas 74(1):79–83. https://doi.org/ trials. BMJ 348(apr01 2):g2035. https://doi.org/10.1136/bmj.g2035 10.1016/j.maturitas.2012.10.008 6. Reid IR, Bolland MJ, Grey A (2014) Effects of vitamin D supple- 22. Bilinski K, Boyages S (2012)The rise andriseofvitamin D ments on bone mineral density: a systematic review and meta-anal- testing. BMJ 345(jul16 2):e4743. https://doi.org/10.1136/bmj. ysis. Lancet 383(9912):146–155. https://doi.org/10.1016/S0140- e4743 6736(13)61647-5 23. Sohl E, Heymans MW, de Jongh RT, den Heijer M, Visser M, 7. DIPART(Vitamin D Individual Patient Analysis of Randomized Merlijn T, Lips P, van Schoor NM (2014) Prediction of vitamin D Trials) Group (2010) Patient level pooled analysis of 68.500 pa- deficiency by simple patient characteristics. Am J Clin Nutr 99(5): tients from seven major vitamin D fracture trials in US and Europe. 1089–1095. https://doi.org/10.3945/ajcn.113.076430 BMJ 340(jan12 1):b5463. https://doi.org/10.1136/bmj.b5463 24. Tran B, Armstrong BK, McGeechan K, Ebeling PR, English DR, 8. Bischoff-Ferrari HA, Willett WC, Orav EJ, Lips P, Meunier PJ, Kimlin MG, Lucas R, van der Pols JC, Venn A, Gebski V, Lyons RA, Flicker L et al (2012) A pooled analysis of vitamin D Whiteman DC, Webb PM, Neale RE (2013) Predicting vitamin D dose requirements for fracture prevention. N Engl J Med 367:481 deficiency in older Australian adults. Clin Endocrinol 79(5):631– 9. Weaver CM, Alexander DD, Boushey CJ, Dawson-Hughes B, 640. https://doi.org/10.1111/cen.12203 Lappe JM, LeBoff MS, Liu S, Looker AC, Wallace TC, Wang 25. Annweiler C, Kabeshova A, Legeay M, Fantino B, Beauchet O DD (2016) Calcium plus vitamin D supplementation and risk of (2015) Derivation and validation of a clinical diagnostic tool for fractures: an updated meta-analysis from the National Osteoporosis the identification of older community-dwellers with Foundation. Osteoporos Int 27(1):367–376. https://doi.org/10. hypovitaminosis D. J Am Med Dir Assoc 16:536.e8–536.19 1007/s00198-015-3386-5 26. Collins GS, Reitsma JB, Altman DG, Moons KG (2015) 10. Avenell A, Mak JC, O’Conell D (2014) Vitamin D and vitamin D Transparent reporting of a multivariable prediction model for indi- analogues for preventing fractures in post-menopausal women and vidual prognosis or diagnosis (TRIPOD): the TRIPOD statement. older men. Cochrane Database Syst Rev 4:CD000227 Ann Intern Med 162(1):55–63. https://doi.org/10.7326/M14-0697 11. Chapuy MC, Arlot ME, Duboeuf F, Brun J, Crouzet B Arnaud S 27. Elders PJM, Merlijn T, Swart KMA, van Hout W, Van der Zwaard (1992) Vitamin D3 and calcium to prevent hip fractures in elderly BC, Niemeijer C et al (2017) Design of the SALT osteoporosis women. N Engl J Med 327(23):1637–1642. https://doi.org/10. study, a randomised pragmatic trial, to study a primary care screen- 1056/NEJM199212033272305 ing and treatment program for the prevention of fractures in women 12. Tricco AC, Thomas SM, Veroniki AA, Hamid JS, Coqo E, Strifler aged 65 years or older. BMC Musculoskelet Disord 18(1):424. L et al (2017) Comparisons of interventions for preventing falls in https://doi.org/10.1186/s12891-017-1783-y older adults: a systematic review and meta-analysis. JAMA 28. Lips P (2012) Interaction between vitamin D and calcium. Scand J 318(17):1687–1699. https://doi.org/10.1001/jama.2017.15006 Clin Lab Invest 243:S60–S64 13. Wu H, Pang Q (2017) The effect of vitamin D and calcium supple- 29. Brot C, Vestergaard P, Kolthoff N, Gram J, Hermann AP, Sorensen mentation on falls in older adults: a systematic review and meta- OH (2001) Vitamin D status and its adequacy in healthy Danish analysis. Orthopade 46(9):729–736. https://doi.org/10.1007/ perimenopausal women: relationships to dietary intake, sun expo- s00132-017-3446-y sure and serum para-thyroid hormone. Br J Nutr 86(S1):S97–103. 14. Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates https://doi.org/10.1079/BJN2001345 S, Clemson LM et al (2012) Interventions for preventing falls in older people living in the community Cochrane Database Syst Rev 30. Farrel C, Soldo J, Williams P, Herrmann M (2012) 25- CD007146 Hydroxyvitamin D-testing: challenging the performance of current 15. Bischoff-Ferrari HA, Dawson-Hughes B, Staehelin HB, Orav JE, automated immunoassays. Clin Chem Lab Med 50:1953–1963 Stuck AE, Theiler R, Wong JB, Egli A, Kiel DP, Henschkowski J 31. Klenk J, Rapp K, Denkinger M, Nagel G, Nikolaus T, Peter R, (2009) Fall prevention with supplemental and active forms of vita- ActiFE Stduy Group et al (2015) Objectively measured physical min D: a meta-analysis of randomized controlled trials. BMJ activity and vitamin D status in older people from Germany. J 339(oct01 1):b3692. https://doi.org/10.1136/bmj.b3692 Epidemiol Community Health 69:388–392 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Osteoporosis International Springer Journals

Prediction of insufficient serum vitamin D status in older women: a validated model

Free
9 pages
Loading next page...
 
/lp/springer_journal/prediction-of-insufficient-serum-vitamin-d-status-in-older-women-a-fFAynTXrOF
Publisher
Springer London
Copyright
Copyright © 2018 by The Author(s)
Subject
Medicine & Public Health; Orthopedics; Endocrinology; Rheumatology
ISSN
0937-941X
eISSN
1433-2965
D.O.I.
10.1007/s00198-018-4410-3
Publisher site
See Article on Publisher Site

Abstract

Summary We developed an externally validated simple prediction model to predict serum 25(OH)D levels < 30, < 40, < 50 and 60 nmol/L in older women with risk factors for fractures. The benefit of the model reduces when a higher 25(OH)D threshold is chosen. Introduction Vitamin D deficiency is associated with increased fracture risk in older persons. General supplementation of all older women with vitamin D could cause medicalization and costs. We developed a clinical model to identify insufficient serum 25-hydroxyvitamin D (25(OH)D) status in older women at risk for fractures. Methods In a sample of 2689 women ≥ 65 years selected from general practices, with at least one risk factor for fractures, a questionnaire was administered and serum 25(OH)D was measured. Multivariable logistic regression models with backward selection were developed to select predictors for insufficient serum 25(OH)D status, using separate thresholds 30, 40, 50 and 60 nmol/L. Internal and external model validations were performed. Results Predictors in the models were as follows: age, BMI, vitamin D supplementation, multivitamin supplementation, calcium supplementation, daily use of margarine, fatty fish ≥ 2×/week, ≥ 1 hours/day outdoors in summer, season of blood sampling, the use of a walking aid and smoking. The AUC was 0.77 for the model using a 30 nmol/L threshold and decreased in the models with higher thresholds to 0.72 for 60 nmol/L. We demonstrate that the model can help to distinguish patients with or without insufficient serum 25(OH)D levels at thresholds of 30 and 40 nmol/L, but not when a threshold of 50 nmol/L is demanded. Conclusions This externally validated model can predict the presence of vitamin D insufficiency in women at risk for fractures. The potential clinical benefit of this tool is highly dependent of the chosen 25(OH)D threshold and decreases when a higher threshold is used. . . . . . Keywords Aged Decision support techniques Logistic models Osteoporosis Vitamin D Vitamin D deficiency * T. Merlijn Introduction tmerlijn@gmail.com Vitamin D supplements are widely used, either self- Department of General Practice and Elderly Care Medicine, administered or with a prescription [1]. The widespread use Amsterdam Public Health Research Institute, VU University Medical is not surprising as vitamin D supplementation for groups at Center, Amsterdam, The Netherlands risk for vitamin D insufficiency is advised in most internation- Stichting ArtsenLaboratorium en Trombosedienst, Koog aan de al and national guidelines. These guidelines find their base in Zaan, The Netherlands the potential harmful effect of low vitamin D status [2–4]. Department of Internal Medicine and Endocrinology, Amsterdam Traditionally, vitamin D deficiency is associated with the dis- Public Health Research Institute, VU University Medical Center, Amsterdam, The Netherlands eases rickets and osteomalacia. Nowadays, these conditions are rare in daily practice. Many vitamin D-related association Department of Health Sciences, Faculty of Earth and Life Sciences, Amsterdam Public Health Research Institute, VU University studies and intervention trials have been conducted in the last Amsterdam, Amsterdam, the Netherlands decades. The observational studies have shown that vitamin D Department of Epidemiology and Biostatistics, Amsterdam Public status is inversely correlated with multiple diseases and con- Health Research Institute, VU University Medical Center, ditions such as fractures, falls, cardiovascular disease, Amsterdam, The Netherlands 1540 Osteoporos Int (2018) 29:1539–1547 different forms of cancer and psychiatric disorders. However, Methods intervention trials have so far only shown positive effects of vitamin D supplementation on bone-related problems and not For the development of our model, we followed the instruc- on other outcomes [5]. Supplementation of vitamin D in com- tion of the TRIPOD [26]. The steps of our method are pre- bination with calcium can reduce loss of bone mineral density sented schematically in Fig. 1. and reduces fracture risk [6–10]. Reduction of fractures by supplementation of both calcium and vitamin D is particularly Study population clear in the oldest group frail persons [11]. Vitamin D with or without calcium supplementation might also reduce falls, but Participants were selected from a randomised prospective trial meta-analyses have shown contradictory results [12–17]. regarding the detection and treatment of older women at risk The serum 25-hydroxyvitamin D (25(OH)D) concentration for fractures in general population and subsequent treatment is the best marker of vitamin D status. The threshold for the with anti-osteoporotic medication, the SALT Osteoporosis need of vitamin D supplementation is still under debate. While Study (SOS) [21]. In this study, women with ages between some advice treatment at 25(OH)D levels of < 25–40 nmol/L, 65 and 90 years were enrolled from GP-practices surrounding others claim that this should be < 75 nmol/L [18, 19]. The five laboratory locations in Noord-Holland, a province in Health Council of the Netherlands considers a level > The Netherlands (Northern Latitude 52°). Women with at least 50 nmol/L sufficient for women older than 50 years and one risk factor for fractures (fracture after the age of 50 years, men older than 70 years and advises daily vitamin D supple- parental hip fracture, low body weight, immobility and con- mentation for this entire group [4]. The Institute of Medicine ditions that may cause secondary osteoporosis) who had been (IOM) has proposed the same threshold. The IOM presumes randomised to the intervention group of the study were eval- that the average requirement of the population is covered with uated with a questionnaire, bone densitometry and laboratory serum 25(OH)D levels of 40 nmol/L [2]. examination [27]. Informed consent was obtained from all When using 50 nmol/L as cut-off, about 50% of the Dutch individual participants and the study was approved by the older persons has a vitamin D insufficiency [20, 21]. Dutch Health Council (OGZ 2.978.265). Considering that only half of all older persons have serum For the development of the prediction model, we used the 25(OH)D levels below the recommended threshold, a general baseline data from all women in the intervention group collected advice to supplement vitamin D in these elderly is only useful in the period from March 2010 until February 2013. An addi- for half of the older population. In the other half of the popu- tional vitamin D-related questionnaire was filled out by the par- lation, treatment would not be necessary, depending on the ticipants that were included during this period. The women who season of blood collection. When lower thresholds are used, had their examination in Amsterdam were excluded because in the number of overtreatment would be even more obvious. this laboratory, a different 25-hydroxyvitamin D assay was used. The consequences are medicalization and costs [22]. Furthermore, we excluded Black, Arabic, Turkish and Asian Testing of serum 25(OH)D levels can reduce the number of women (n = 103) and women in a residential care home (n = unnecessary treatments but is also expensive and the outcome is 24), resulting in a total of 2689 participants. The reason for ex- influenced by the period of the year. Three prediction models to cluding these women was that there were too few participants in predict vitamin D deficiency in elderly were published before these groups and most of them already had an indication for [23–25]. In an analogous manner, we wanted to develop a pre- vitamin D supplementation. diction model to predict insufficient vitamin D status in older For the external validation of the prediction model, we collect- women. In contrast to previous studies, our study focuses on ed data from a second population in the same region. All women the clinical implementation of the model. Therefore, we focused between 65 and 90 years old who had been referred by a general on older women with risk factors for fractures and evaluated practitioner for combined dual X-ray absorptiometry (DXA) and clinical applicability. In the second place, we used a spectrum laboratory examination from September 2010 until February of thresholds, in a way that medical professionals can choose the 2013 were asked to participate and to complete the same ques- threshold they consider to be appropriate. tionnaires (n = 856). This was a non-overlapping population in The goal of this study was to develop a validated prediction the same region that was evaluated in the same laboratory. model that could discriminate older women who do or do not need treatment with vitamin D supplements. We aimed to Questionnaires construct a simple tool that can be added to fracture risk eval- uation in general practice or that can be used as a self-test. All women completed two questionnaires. The first question- While there is still discussion about the optimal cuff of value naire contained questions about risk factors for fractures and of 25(OH)D levels, we will develop the model with different osteoporosis. If a questionnaire was returned incomplete, the thresholds for serum 25(OH)D levels, namely 30, 40, 50 and participant was contacted by telephone to complete the miss- 60 nmol/L. ing answers. The second questionnaire contained questions Osteoporos Int (2018) 29:1539–1547 1541 Fig. 1 Scheme of the Populaon women ≤65 years development and validation of the prediction model ≥1 risk factor for fractures Quesonnaire, biometry, and serum 25(OH)D Exclusion: -Women in nursing home - Black, Arabic and Asian women Data exploraon/ preparaon: -Choosing cut-offs in connuous variables -Opmalizaon modeling of the seasons -Imputaon if more than 5% missings Predicon models: logisc regression with backward selecon (p<0.157) Four thresholds: 30, 40, 50, and 60 nmol/L Internal validaon of the 4 models Simplificaon of regression coefficients to 4 risks scores External validaon primary models and risk scores Calculaon of predicve values of models and examples of praccal use about predictors of vitamin D status. The questionnaires were immobility, use of walking aid, falling during previous collected at the laboratory visit when DXA and blood tests 12 months, vitamin D supplementation either self- were performed. administered or prescribed, multivitamin supplementation, calcium supplementation, smoking, consumption of fatty fish, Laboratory analyses consumption of margarine, alcohol use, time spending outside in winter and summer, the level of education and Serum 25(OH)D was analysed with a chemoluminescence polypharmacy. assay (Diasorin, Stillwater, MN, USA) as a routine measure- Age was calculated as date of examination minus date of ment at the Centrum voor Medische Analyse Antwerpen. The birth. Weight and height were measured during the visit of the laboratory is accredited and meets the European norm ISO laboratory and BMI was calculated as kg/m . 15180. The inter-assay coeficient of variation was 10.1% at All other questions were self-reported. Patients were con- a 25(OH)D concentration of 61.5 nmol/L and 9.9% at a sidered to be immobile when severe problems with walking 25(OH)D concentrations of 36 nmol/L. were indicated (yes/no). Use of walking aid, current smoking, falls (at least 1 in the last 12 months), supplementation of Outcome vitamin D, calcium and multivitamin use (daily) were dichot- omous questions (yes/no). In a second question, the partici- In the models, we used a dichtome outcome: below and equal pants were asked to specify whether the supplementation of to or above the threshold level of 25(OH)D of, respectively, vitamin D was on prescription or self-administered. 30, 40, 50 and 60 nmol/L. Time spent outdoors in winter and summer, level of edu- cation, alcohol use, fatty fish and margarine use were mea- Predictors of vitamin D status sured in four categories. For the model, these categories were dichotomized. The selected potential predictors were age and body mass Medication use was collected with the question to list all actual index (BMI), period of the year of blood sampling, medication. We defined polypharmacy as the use of six drugs or 1542 Osteoporos Int (2018) 29:1539–1547 more. We used individual dates of blood samples for serum For the external validation, we calculated the AUC of 25(OH) measurement to determinate the period of the year. models and the practical model based on the risk score in a second population. Model development Examples of application The development of the prediction model was preceded by the We shall show two examples how the prediction model can be preparation of some potential predictors. Before modelling, we used for a practical purpose. In the first place, we will produce evaluated collinearity of the variables. There was none. We used figures with predictive values for the different thresholds. That splines to evaluate non-linear relationships between continuous might be useful for predicting the probability for deficiency in predictors and the outcome and to modify continuous variables daily practice. In the second place, we apply this tool to esti- into categories. In order to reduce the variables for the period of mate the proportion of women at risk for 25(OH) insufficiency the year, we observed the pattern of seasonal change of 25(OH)D that need supplementation in a specific period of the year. levels over the months during 3 years. We used univariate linear regression of 25(OH)D levels to find an optimal reflection of the Software seasonal influence in a maximum of four periods and selected the model with the highest R and the least amount of periods. IBM SPSS Statistics version 22 was used for data checking To select the predictors for the model, we performed logistic and modelling and R 3.2.2 was used for bootstrapping and regression analyses. We used a backward selection strategy. That assessment of the final models. means that we started with all potential predictors in the model, and every turn, the predictor with the highest P value was ex- cluded from the model, until all P values were lower than our Results selection criterion of p<0.157 (Akaike’s information criterion). A stricter p value leads to the development of models that are Population characteristics closely adapted to the data and that generalise poorly. Since missing data can effect a prognostic model, we Between March 2010 and February 2013, we collected ques- planned to use multiple imputation to estimate the missing tionnaires and serum 25(OH)D values of 2689 women. Of values according to the Multivariate Imputation by Chained these women, 2624 (97.6%) had complete data; hence, impu- Equation procedure in IBM SPSS Statistics 20, only in case of tation was not necessary. Serum 25(OH)D levels ≥ 60, ≥ 50, ≥ more than 5% missing cases. Due to few missing cases, im- 40 and ≥ 30 nmol/L were found in 32, 50, 65 and 90% of the putation was not necessary. women, respectively. The population for external validation comprised 856 women. The main characteristics of both pop- Model performance ulations are shown in Table 1. To evaluate the discrimination of the model, i.e., if the model Modelling of seasonal changes is able to distinguish patients with and without an insufficient 25(OH)D status, a receiver operator characteristic (ROC) Figure 2 shows the seasonal change of the mean serum curve was made and the area under the roc curve (AUC) 25(OH)D levels. The model with the highest R and the lowest calculated. The goodness-of-fit of the model was tested by number of periods was a model with three variables: the nadir the Hosmer-Lemeshow test. The unexplained variance was in December till April, a peak in July and August and inter- indicated by Nagelkerke’s R . mediate levels in the rest of the months. Internal validation was performed with bootstrapping tech- niques. Regression coefficients and performance of the Development of the prediction models models were adjusted according to the optimism estimates from the internal validation procedure. Irrespective of the chosen threshold, the best prediction model We constructed a risk score by multiplying the regression contained the following predictors: age, BMI, walking aid, coefficients of the predictors by 10 and divided by 3. The vitamin D supplementation either self-administered or pre- result was rounded to the nearest whole number. The last step scribed, multivitamin use, calcium supplementation, smoking, was made to keep the scores low to keep the model practical. time spent outdoors in summer and period of blood sample. At To check the loss of information, we compared AUC before thresholds of 30 and 40 nmol/L, use of margarine and fatty and after dividing by 3. We calculated the sensitivity, speci- fish were predictors. From a threshold of 50 nmol/L, use of ficity, positive predictive value and negative predicted value fatty fish was not a predictor and use of margarine was a very for the defined thresholds. weak predictor and disappeared from the model at higher Osteoporos Int (2018) 29:1539–1547 1543 Table 1 The prevalence of the determinant in the study population, the regression coefficients of the models with different cut-offs and corresponding risk scores Model 1: threshold Model 2: threshold Model 3: threshold Model 4: threshold 30 nmol/L 40 nmol/L 50 nmol/L 60 nmol/L Primary population External sample Regression Risk Regression Risk Regression Risk Regression Risk N = 2689 N = 856 coefficient (B) score coefficient (B) score coefficient (B) score coefficient (B) score 1 2 3 4 25-hydroxyvitamin D (nmol/L) 51.8 (20.5) 54.7(20.9) < 30 14.1% 11.3% < 40 30.7% 27.6% < 50 50.4% 54.3% < 60 68.1% 60.4% Age (years) 73.5(6.1) 73.1 (6.0) 65–70 32.6% 34.2% 70–75 27.3% 29.0% 0.30 1 0.32 1 0.34 1 > 75 40.0% 36.8% 0.59 2 0.75 2 0.69 2 0.67 2 Body mass index (kg/m ) 28.0 (6.1) 27.1 (4.5) < 25 28.7% 33.5% 25–30 40.9% 41.0% 0.27* 1 0.22 1 0.27 1 > 30 30.4% 25.5% 0.44 1 0.61 2 0.52 2 0.44 1 Impaired mobility (% yes) 26.0% 17.4% 0.24* 1 Walking aid (% yes) 26.2% 18.3% 0.63 2 0.58 2 0.48 2 0.45 1 No of falls in last year (% ≥ 1) 29.9% 46.0% 0.20 1 0.15* 1 Vitamin D supplementation (% 17.3% 30.4% yes) Self-administered 11.0% 15.4% − 0.98 − 3 − 0.72 − 2 − 0.74 − 2 − 0.41 − 1 Prescribed 6.3% 15.0% − 0.88 − 3 − 1.36 − 5 − 1.55 − 5 − 1.16 − 4 Use of multivitamins (% yes) 24.4% 23.4% − 0.52 − 2 − 0.79 − 3 − 0.92 − 3 − 0.65 − 2 Calcium supplementation (% yes) 19.7% 29.1% − 0.38 − 1 − 0.44 − 1 − 0.52 − 2 − 0.45 − 1 Fattyfishconsumption(% 12.7% 12.1% − 0.43 − 1 − 0.27 − 1 ≥ 2×/week) Margarine consumption daily (% 69.6% 69.2% − 0.54 − 2 − 0.32 − 1 − 0.15* − 1 yes) Smoking (% yes) 10.2% 11.9% 0.52 2 0.30* 1 0.30 1 0.21* 1 Use of alcohol (% ≥ 1 unit/day) 25.7% 27.5% Time outdoors in winter (% 22.9% 21.7% − 0.23* − 1 − 0.24 − 1 > 60 min/day) Time outdoors in summer (% 80.3% 77.6% − 0.67 − 2 − 0.63 − 2 − 0.50 − 2 − 0.36 − 1 > 60 min/day) Month of examination (%) Dec-Apr 34.0% 45.1% 1.62 5 1.50 5 1.26 4 1.19 4 May-Jun or Sep-Nov 42.8% 42.1% 0.83 3 0.79 3 0.70 2 0.63 2 Jul-Aug 23.3% 12.8% Fracture > 50 year of age (% yes) 45.6% 55.1% Parental hip fracture (% yes) 26.9% 13.6% No of medication (% > 6) 17.5% 15.4% Education (% none or low) 80.0% 79.5% Constant − 2.53 − 1.52 − 0.56 0.18 *Significance level of p < 0.154 All other regression coefficients significance level of p < 0.05 1544 Osteoporos Int (2018) 29:1539–1547 sensitivity, specificity, positive predictive value and neg- ative predictive value of the model. The risk scores for the different models are shown in Table 1. Application of the risk score To use the model, one has to choose a threshold. Figure 3 shows the predictive value for the different thresholds. For example, when a threshold of 50 nmol/ L is desired and a patient has a risk score of 0 or lower, this patient has a 70% probability to have 25(OH)D levels above the threshold. Table 3 shows the proportion women at risk for 25 (OH)D insufficiency in the primary population without vitamin D supplementation. The proportion is presented for the different thresholds, a desired probability and per season. For example, to have a probability of 80% of having 25(OH)D levels above 50 nmol/L, 100% are at Fig. 2 Mean serum 25(OH)D levels per month. Error bars show 95% CI risk in winter, 94% in spring and autumn and 79% in of the mean summer. This shows that the model could only differen- tiate few women who do not need treatment at this thresholds. Falling on the other hand was a predictor at the threshold. However, at a threshold of 40 nmol/L, the higher thresholds but not at 30 nmol/L. Time spent outdoor in model selects 88% at risk in winter and 50% in spring winter was a predictor at thresholds of 40 and 50 nmol/L. and autumn. What means that the model is able to se- Table 1 shows the β’s of the different predictors at the differ- lect 50% of the population that would not need treat- ent thresholds. ment most of the year. It is clear that at a threshold of 30 nmol/L, even more women not at risk could be se- Validation of the prediction model lected but nihil at a threshold of 60 nmol/L. The internally validated models showed an AUC of 0.72 to 0.77 with the highest for the threshold 30 nmol/L. There is little loss of information when the model is Discussion converted to a risk score. The R is 0.25 for the thresh- old of 40 nmol/L. The AUC of the external validation In this study, we present four validated models for the predic- was between 0.71 and 0.82 (Table 2). Figure 3 shows the tion of vitamin D status for different 25(OH)D thresholds Table 2 AUCofthe ROCcurve AUC after Nagelkerke AUC in and Nagelkerke R square of the internal R square external internal validated models and the validation sample simplified models for the different 25(OH)D thresholds Model threshold 30 nmol/L Prediction model 0.77 0.21 0.82 Risk score 0.77 0.21 0.82 Model threshold 40 nmol/L Prediction model 0.76 0.25 0.75 Risk score 0.76 0.25 0.74 Model threshold 50 nmol/L Prediction model 0.75 0.24 0.72 Risk score 0.75 0.24 0.72 Model threshold 60 nmol/L Prediction model 0.73 0.18 0.71 Risk score 0.72 0.18 0.71 AUC area under the curve, ROC receiver operator characteristics For the simplified models, the regression coefficients were multiplied by 3/10 and rounded to the nearest whole number Osteoporos Int (2018) 29:1539–1547 1545 Model 1: 25(OH)D ≤ 30nmol/L Model 2: 25(OH)D ≤ 40nmol/L 100% 100% 90% 90% 80% 80% 70% 70% 60% 60% 50% 50% 40% 40% 30% 30% 20% 20% 10% 10% 0% 0% -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 Model 4: 25(OH)D ≤ 60nmol/L Model 3: 25(OH)D ≤ 50nmol/L 100% 100% 90% 90% 80% 80% PPV 70% 70% NPV 60% 60% 50% 50% Specificity 40% 40% 30% 30% Sensivity 20% 20% Prevalance 10% 10% 0% 0% -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Fig. 3 Four models with different thresholds of serum 25-hydroxyvitamin D. The Y-axis shows the positive predictive (PPV), negative predictive value (NPV), sensitivity, specificity and prevalence for any computed risk score (X-axis) of participants in primary population between 30 and 60 nmol/L. In our primary sample and in an 25(OH)D was tested. Some predictors were not in all models. external sample, reasonable predictive values were shown. The intake of fatty fish is only present in the models with The evaluation of the models with different thresholds eluci- threshold 30 and 40 nmol/L, respectively. The use of marga- dates that the model with a threshold of 60 nmol/L has less rine is a predictor in the models with thresholds from 30 to predictors and lower prediction than the other models. 50 nmol/L, but is a weak predictor at 50 nmol/L. This suggest Furthermore, we have shown that the models with thresholds that food-related vitamin D intake helps to prevent the severest of 50 and 60 nmol/L have low negative predictive value and deficiency, but not to maintain higher values. therefore are less potent to exclude low 25(OH)D levels in our Most predictors are similar to those observed in earlier population, due to the high prevalence of 25(OH)D levels studies. The AUC in other studies for a threshold of lower than these thresholds. The highest AUC were seen at 50 nmol/L range from 0.73 to 0.86 [17–19]. Only one study thresholds of 30 and 40 nmol/L. (AUC of 0.71) was validated in an external population [17]. Predictors for insufficient 25(OH)D in all models were The age of the participants in these studies was similar, but higher age, higher BMI, use of walking aid, little time spent there was difference in the prevalence of 25(OH)D deficiency. outdoors in summer, smoking, no calcium supplementation, In contrast to our model, other studies added more complex no use of multivitamins, no use of vitamin D on prescription predictors that consist of more than one question or a physical or self-administered and the period of the year when serum examination. Table 3 Application of the models: percentage of women at risk for 25(OH)D insufficiency per season for different thresholds of 25(OH)D and probability to reach the threshold Threshold 25(OH)D Model 1 30 nmol/L Model 2 40 nmol/L Model 3 50 nmol/L Model 4 60 nmol/L Percentage above threshold 70% 80% 90% 70% 80% 90% 70% 80% 90% 70% 80% 90% 25(OH)D insufficiency Not at all 100% 84% 20% 35% 12% 1% 3% 0% 0% 0% 0% 0% Winter 0% 16% 80% 65% 88% 99% 97% 100% 100% 100% 100% 100% Spring and autumn 0% 0% 18% 4% 50% 89% 79% 94% 100% 99% 100% 100% Summer 0% 0% 0% 0% 0% 51% 27% 79% 100% 99% 99% 99% Calculated in the participants without vitamin D supplementation in the primary population 1546 Osteoporos Int (2018) 29:1539–1547 We have observed that the use of calcium supplements is simple measurement that lacks in our model that might improve associated with higher 25(OH)D levels, which can be explained the performance of the model is physical activity [23, 24]. by a lower vitamin D metabolite turnover due to reduction of the Information on physical activity was not available in our study, parathyroid hormone levels [28]. but there may be an overlap with impaired mobility, the use of a With respect to vitamin D supplementation, it is remarkable walking aid and the time that is spent outdoors [23]. Time spent that vitamin D on prescription is a stronger predictor than self- outdoors in summer as well in winter are positive predictors. administered vitamin D. There are two possible explanations. Because of the low intensity of the sun in winter, we do not think First, the compliance might be better when vitamin D is given that this is a direct effect, but that it might be a reflection of on prescription. Secondly, it is possible that patients obtain physical activity and general health as well [31]. lower dosages without prescription. The 25 (OH)D assay we used was not compared with the The period of the year is one of the strongest predictors in our reference of the national institute of standards and technology. model. As our data covers 3 years, we observed that there were Nevertheless, we used a single 25(OH)D assay with good perfor- substantial differences per year. In most studies, the change in mance [30]. serum 25(OH)D levels is supposed to be a symmetric sinusoidal In conclusion, the proposed model can help to distinguish curve over the seasons. However, in our data, the mean levels women with adequate serum 25(OH)D levels in a population of over 3 years show a pattern in which the rise of serum 25(OH)D older women with risk factors for fractures. This might reduce in spring is steeper than the decrease in autumn, a narrow peak in unnecessary treatment with vitamin D in some relatively healthy summer and a wide dip in winter. The result of this pattern is that older women. To reach levels above the 50 nmol/L during the the optimal model does not exactly follow the seasons. The peak whole year, practically all women need supplementation and is in the months of July and August; the dip is from December many of them need continuous supplementation. If one allows until the end of April. The other months are between these sum- levels of 30 or 40 nmol/L, the model can substantially reduce the mer and winter levels. A similar pattern has been reported earlier number of women that need supplementation, at least during a in a Danish cohort of postmenopausal women [29]. part of the year. This shows that advices in guidelines can have A strength of this study is that it was performed in a large substantial impact on the number of women with an indication for population of women at risk for fractures and that it was val- supplementation. idated in an external population of women with comparable Acknowledgements The SALT Osteoporosis Study has been largely age. In the external validation, the loss of discrimination was funded by Stichting Achmea Gezondheidszorg. Healthcare costs have limited or absent, indicating that the model is robust. been compensated by Achmea and VGZ Zorgverzekeraar. Additional There are different opinions about the optimal threshold for financial support has been provided by Stichting ArtsenLaboratorium serum 25(OH)D [18, 19]. Also, one might differ about wheth- en Trombosedienst. The sponsors do not have any role in the design or implementation of the study, data collection, data management, future er one prefers a high specificity with a high positive predictive data analysis and interpretation, or in the preparation, review or approval value, but more false negatives or one prefers a high sensitiv- of the manuscript. ity with a high negative predictive value but more false posi- tives. Therefore, we have presented our data in a way that Compliance with Ethical Standards medical professionals can choose between different thresholds and between different positive and negative predictive values. Informed consent was obtained from all individual participants and the study The selection of a population of women with increased was approved by the Dutch Health Council (OGZ 2.978.265). fracture risk has advantages and disadvantages. A profit is that Conflicts of interest The authors declare that they have no conflict of we have exact insight in the performance of the model in a interest. population in which 25(OH)D levels are clinically most im- portant. A disadvantage is that the application of the model Ethical approval All procedures performed in studies involving human has only been validated in Dutch Caucasian women with in- participants were in accordance with the ethical standards of the institu- creased fracture risk, and therefore, we must be careful with a tional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. broader implementation of this model. Unfortunately, we did not have enough data of non-Caucasian women living in the Open Access This article is distributed under the terms of the Creative Netherlands to gain more insight in these specific groups. Commons Attribution-NonCommercial 4.0 International License (http:// creativecommons.org/licenses/by-nc/4.0/), which permits any noncom- Another limitation is the lack of information on the dose of mercial use, distribution, and reproduction in any medium, provided vitamin D supplementation. Furthermore, a large part of variance you give appropriate credit to the original author(s) and the source, pro- is not explained. The unexplained variance might be reduced by vide a link to the Creative Commons license, and indicate if changes were improvement of the accuracy of the measurements, e.g., with the made. use of standardised intake questionnaires or the measurement of ultraviolet exposure. However, for our purpose, a simple tool for daily practice, more complex measurements are not suitable. One Osteoporos Int (2018) 29:1539–1547 1547 16. Bolland MJ, Grey A, Reid IR (2014) Differences in overlapping References meta-analysis of vitamin D supplements and falls. J Clin Endocinol Metab 99(11):4265–4272. https://doi.org/10.1210/jc.2014-2562 1. Kantor ED, Rehm CD, DU M, White E, Giovannucci EL (2016) 17. Murad MH, Elamin KB, Elnour OA, Elamin MB, Alkatib AA, Trends in dietary supplement use among US adults form 1999- Fatourechi MM et al (2011) The effect of vitamin D on falls: a 2012. JAMA 316(14):1464–1474. https://doi.org/10.1001/jama. systematic review and meta-analysis. J Clin Endocrinol Metab 2016.14403 96(10):2997–3006. https://doi.org/10.1210/jc.2011-1193 2. Ross AC, Manson JE, Shapses SA (2011) The 2011 report on 18. Reid IR (2014) Skeletal and nonskeletal effects of vitamin D: is dietary reference intakes for calcium and vitamin D from the insti- vitamin D a tonic for bone and other tissues? Osteoporos Int tute of medicine: what clinicians need to know. J Clin Endocrinol 25(10):2347–2357. https://doi.org/10.1007/s00198-014-2749-7 Metab 96(1):53–58. https://doi.org/10.1210/jc.2010-2704 19. Bischoff-Ferrari HA (2010) Vitamin D and fracture prevention. 3. Wood CL, Cheetham TD (2016) Vitamin D: increasing use among Endocrinol Metab Clin N Am 39(2):347–353. https://doi.org/10. at-risk groups (NICE guideline PH56). Arch Dis Child Educ Ed 1016/j.ecl.2010.02.009 101(1):43–45. https://doi.org/10.1136/archdischild-2015-308299 20. Van Schoor NM, Visser M, Pluijm SM, Kuchuk N, Smit JH, Lips P 4. Health Council of the Netherlands (2012) Evaluation of the dietary (2008) Vitamin D deficiency as a risk factor for osteoporotic frac- reference values for vitamin D. Health Council of the Netherlands, tures. Bone 42(2):260–266. https://doi.org/10.1016/j.bone.2007.11. The Hague 5. Theodoratou E, Tzoulaki I, Zgaga L, Ioannidis JP (2014) Vitamin D 21. Janssen HCJP, Emmelot-Vonk MH, Verhaar HJJ, van der Schouw and multiple health outcomes: umbrella review of systematic re- YT (2013) Determinants of vitamin D status in healthy men and views and meta-analyses of observational studies and randomised women aged 40-80 years. Maturitas 74(1):79–83. https://doi.org/ trials. BMJ 348(apr01 2):g2035. https://doi.org/10.1136/bmj.g2035 10.1016/j.maturitas.2012.10.008 6. Reid IR, Bolland MJ, Grey A (2014) Effects of vitamin D supple- 22. Bilinski K, Boyages S (2012)The rise andriseofvitamin D ments on bone mineral density: a systematic review and meta-anal- testing. BMJ 345(jul16 2):e4743. https://doi.org/10.1136/bmj. ysis. Lancet 383(9912):146–155. https://doi.org/10.1016/S0140- e4743 6736(13)61647-5 23. Sohl E, Heymans MW, de Jongh RT, den Heijer M, Visser M, 7. DIPART(Vitamin D Individual Patient Analysis of Randomized Merlijn T, Lips P, van Schoor NM (2014) Prediction of vitamin D Trials) Group (2010) Patient level pooled analysis of 68.500 pa- deficiency by simple patient characteristics. Am J Clin Nutr 99(5): tients from seven major vitamin D fracture trials in US and Europe. 1089–1095. https://doi.org/10.3945/ajcn.113.076430 BMJ 340(jan12 1):b5463. https://doi.org/10.1136/bmj.b5463 24. Tran B, Armstrong BK, McGeechan K, Ebeling PR, English DR, 8. Bischoff-Ferrari HA, Willett WC, Orav EJ, Lips P, Meunier PJ, Kimlin MG, Lucas R, van der Pols JC, Venn A, Gebski V, Lyons RA, Flicker L et al (2012) A pooled analysis of vitamin D Whiteman DC, Webb PM, Neale RE (2013) Predicting vitamin D dose requirements for fracture prevention. N Engl J Med 367:481 deficiency in older Australian adults. Clin Endocrinol 79(5):631– 9. Weaver CM, Alexander DD, Boushey CJ, Dawson-Hughes B, 640. https://doi.org/10.1111/cen.12203 Lappe JM, LeBoff MS, Liu S, Looker AC, Wallace TC, Wang 25. Annweiler C, Kabeshova A, Legeay M, Fantino B, Beauchet O DD (2016) Calcium plus vitamin D supplementation and risk of (2015) Derivation and validation of a clinical diagnostic tool for fractures: an updated meta-analysis from the National Osteoporosis the identification of older community-dwellers with Foundation. Osteoporos Int 27(1):367–376. https://doi.org/10. hypovitaminosis D. J Am Med Dir Assoc 16:536.e8–536.19 1007/s00198-015-3386-5 26. Collins GS, Reitsma JB, Altman DG, Moons KG (2015) 10. Avenell A, Mak JC, O’Conell D (2014) Vitamin D and vitamin D Transparent reporting of a multivariable prediction model for indi- analogues for preventing fractures in post-menopausal women and vidual prognosis or diagnosis (TRIPOD): the TRIPOD statement. older men. Cochrane Database Syst Rev 4:CD000227 Ann Intern Med 162(1):55–63. https://doi.org/10.7326/M14-0697 11. Chapuy MC, Arlot ME, Duboeuf F, Brun J, Crouzet B Arnaud S 27. Elders PJM, Merlijn T, Swart KMA, van Hout W, Van der Zwaard (1992) Vitamin D3 and calcium to prevent hip fractures in elderly BC, Niemeijer C et al (2017) Design of the SALT osteoporosis women. N Engl J Med 327(23):1637–1642. https://doi.org/10. study, a randomised pragmatic trial, to study a primary care screen- 1056/NEJM199212033272305 ing and treatment program for the prevention of fractures in women 12. Tricco AC, Thomas SM, Veroniki AA, Hamid JS, Coqo E, Strifler aged 65 years or older. BMC Musculoskelet Disord 18(1):424. L et al (2017) Comparisons of interventions for preventing falls in https://doi.org/10.1186/s12891-017-1783-y older adults: a systematic review and meta-analysis. JAMA 28. Lips P (2012) Interaction between vitamin D and calcium. Scand J 318(17):1687–1699. https://doi.org/10.1001/jama.2017.15006 Clin Lab Invest 243:S60–S64 13. Wu H, Pang Q (2017) The effect of vitamin D and calcium supple- 29. Brot C, Vestergaard P, Kolthoff N, Gram J, Hermann AP, Sorensen mentation on falls in older adults: a systematic review and meta- OH (2001) Vitamin D status and its adequacy in healthy Danish analysis. Orthopade 46(9):729–736. https://doi.org/10.1007/ perimenopausal women: relationships to dietary intake, sun expo- s00132-017-3446-y sure and serum para-thyroid hormone. Br J Nutr 86(S1):S97–103. 14. Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates https://doi.org/10.1079/BJN2001345 S, Clemson LM et al (2012) Interventions for preventing falls in older people living in the community Cochrane Database Syst Rev 30. Farrel C, Soldo J, Williams P, Herrmann M (2012) 25- CD007146 Hydroxyvitamin D-testing: challenging the performance of current 15. Bischoff-Ferrari HA, Dawson-Hughes B, Staehelin HB, Orav JE, automated immunoassays. Clin Chem Lab Med 50:1953–1963 Stuck AE, Theiler R, Wong JB, Egli A, Kiel DP, Henschkowski J 31. Klenk J, Rapp K, Denkinger M, Nagel G, Nikolaus T, Peter R, (2009) Fall prevention with supplemental and active forms of vita- ActiFE Stduy Group et al (2015) Objectively measured physical min D: a meta-analysis of randomized controlled trials. BMJ activity and vitamin D status in older people from Germany. J 339(oct01 1):b3692. https://doi.org/10.1136/bmj.b3692 Epidemiol Community Health 69:388–392

Journal

Osteoporosis InternationalSpringer Journals

Published: May 28, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off