Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Predicting Chronic Obstructive Pulmonary Disease Hospitalizations Based on Concurrent Influenza Activity

Predicting Chronic Obstructive Pulmonary Disease Hospitalizations Based on Concurrent Influenza... 2 University of Iowa Department of Biostatistics, Iowa City, IA, USA Although infl uenza has been associated with chronic obstructive pulmonary 3 Harvard School of Public Health: 651 Huntington disease (COPD) exacerbations, it is not clear the extent to which this association Avenue, Boston, MA, USA affects healthcare use in the United States. The fi rst goal of this project was to determine to what extent the incidence of COPD hospitalizations is associated with seasonal infl uenza. Second, as a natural experiment, we used infl uenza activity to help predict COPD admissions during the 2009 H1N1 infl uenza pandemic. To do this, we identifi ed all hospitalizations between 1998 and 2010 in the Nationwide Inpatient Sample from the Healthcare Cost and Utilization Project (HCUP) during which a primary diagnosis of COPD was recorded. Separately, we identifi ed all hospitalizations during which a diagnosis of infl uenza was recorded. We formulated time series regression models to investigate the association of monthly COPD admissions with infl uenza incidence. Finally, we applied these models, fi t using 1998–2008 data, to forecast monthly COPD admissions during the 2009 pandemic. Based on time series regression models, a strong, signifi cant association exists between concurrent infl uenza activity and incidence of COPD hospitalizations (p-value < 0.0001). The association is especially strong among older patients requiring mechanical ventilation. Use of infl uenza data to predict COPD admissions during the 2009 H1N1 pandemic reduced the mean-squared prediction error by 29.9%. We conclude that infl uenza activity is signifi cantly associated with COPD hospitalizations in the United States and infl uenza activity can be exploited to more accurately forecast COPD admissions. Our results suggest that improvements in infl uenza surveillance, prevention, and treatment may decrease hospitalizations of patients diagnosed with COPD. Introduction Chronic obstructive pulmonary disease (COPD) is a major cause of morbid- f morbid- bi id- d is one of t o of t f th he e ity and mortality in the United States (1). In addition, COPD is one of the major or c r com o ompo po- o- - most expensive chronic diseases, and hospitalizations are a major compo- cute exac xac ac ce e erb rb ba atio t tions ion nent of direct medical costs attributable to COPD (2, 3). Acute exacerbations matory ry e e ep ep piso i iso isod de de es sst s tr t ig- ig g g g- of COPD (AECOPD) are thought to be acute infl ammatory episodes trig- Keywords: Chronic airfl ow obstruction, human infl uenza, epidemiology, hospitalization, disease pathogens ge ens, v , i ir r ru u us use ses s,, and a an and nd gered by environmental factors, including bacterial pathogens, viruses, and exacerbation, chronic obstructive pulmonary disease, en infl uenz uen nz nza a a and AEC an and nd dA AEC EC CO OPD O OPD D ambient air pollution (4, 5). Th e association between infl uenza and AECOPD COPD exacerbation. g epithe he el l lial c ial ia ce el lls and r lsa s nd d r re e e es sult sult ta an ant t is thought to be mediated via destruction of lung epithelial cells and resultant increa as s se e ed dp dp d prr ro o oduc du d duc u tion of m io on o mu m mu u uc cu u us s s, ,, infl ammation leading to lung tissue damage, increased production of mucus, Correspondence to: Philip M. Polgreen, MD, n. ThThe e e e e ex x xac a t d t dis di i ea as se e me em e ec chanisms ch ch nism sm m vasoconstriction, and bronchoconstriction. Th e exact disease mechanisms University of Iowa Department of Internal Medicine: 200 Hawkins Dr., Iowa City, Iowa 52242, er, episo iso od d de es of r of f r f r re e espira s spir pira ra at tor ry yyf f f failur a lur urea e e and nd d d are not completely understood. However, episodes of respiratory failure and phone: 319-356-1616, fax: 319-353-6406, sociate ed ed d w d w d w wi it i it th h ac h ac ac acc ce c celera era ra ate ted de d d de decline of lung ine f lu lu un n mechanical ventilation have been associated with accelerated decline of lung email; [email protected] n patien nt ts ts s diag d dia ag gn gn no nose e ed d w dw w wi it th h h C h C C CO OPD (6, 7). PD PD D D(6, 7 7 7) function and poor quality of life in patients diagnosed with COPD (6, 7). 573 573 574 Gerke et al. Epidemiologically, the seasonality of infl uenza infec- a subgroup of patients admitted specifi cally for acute tions and COPD exacerbations is similar, with both COPD exacerbations (ICD-9-CM code: 491.21). peaking during winter months from December to Feb- For both of these groups, we then identifi ed patients ruary (8). Th us, although infl uenza has been associated who required mechanical ventilation using the ICD-9- with COPD exacerbations, it is less clear how important CM codes 93.90 (non-invasive mechanical ventilation) these patterns are regarding healthcare use. Under- and/or 96.70–96.72 (continuous invasive mechanical standing the extent of the association between infl uenza ventilation). Separately, to build our infl uenza series, and COPD hospitalizations could lead to more focused we identifi ed all hospitalizations from January 1998 and cost-eff ective eff orts to prevent hospitalizations and through July 2010 during which a primary or secondary develop novel interventions. Predicting and quantifying diagnosis of infl uenza was received. For infl uenza case the burden of COPD hospitalizations using infl uenza- ascertainment, we used the ICD-9-CM codes 487.00 related data streams may also allow hospitals to plan for (infl uenza with pneumonia), 487.10 (infl uenza with increases in hospital resources (e.g. pharmaceuticals, other respiratory manifestations), and 487.80 (infl uenza ventilators, and staffi ng of critical care units). with other manifestations). We then compiled monthly Th e purpose of this study is to use time series meth- totals of cases of infl uenza in the same manner as COPD ods to determine whether, and to what extent, the inci- based on the month that the patient was admitted. Using dence of COPD hospitalizations is associated with the this same approach, we compiled COPD and infl uenza seasonal variation in the incidence of infl uenza. Because series from the four census regions (Northeast, Midwest, both infl uenza and COPD may be caused by some other South, and West) to confi rm the results at the national seasonal and winter-related factor, we use the unusu- level. ally early peak of the 2009 H1N1 infl uenza pandemic in Time series correlation analysis October as a natural experiment to determine if infl u- To investigate the association of COPD with infl uenza, enza and COPD admissions are associated outside of the traditional infl uenza season typically comprised of we computed cross-correlation functions (CCF) for winter months (December through February). Finally, the infl uenza series and each of our four COPD series: we provide estimates of a measure of attributable risk COPD, AECOPD, COPD with mechanical ventilation, that further characterize the relationship between and AECOPD with mechanical ventilation. Th e CCF COPD and infl uenza. indicates the temporal correlations between two time series: specifi cally, a series at time t and another series at time t+m, where m is referred to as the lag. Because cross- Methods correlations between time series can be spurious due to Data source the eff ects of common temporal patterns, we employed All data were extracted from the Nationwide Inpatient a prewhitening process (10). In our application, com- Sample (NIS), the largest all-payer database of national mon yearly cycles are present in both the COPD series discharges in the U.S. Th e database is maintained as and the infl uenza series, since both are elevated during part of the Healthcare Cost and Utilization Project the winter months. Th e prewhitening process allows us (HCUP) by the Agency for Healthcare Research and to detect correlations based on prominent local peaks or Quality (AHRQ), and contains data from a 20% strati- troughs in two time series that are temporally aligned, fi ed sample of nonfederal acute care hospitals (9). Th is as opposed to coincidental correlations based on shared sample includes academic medical centers, community seasonal patterns. Th e former are representative of a hospitals, general hospitals, and specialty hospitals. legitimate association, whereas the latter are merely due It excludes long-term care facilities and rehabilitation to common cyclic behavior. hospitals. To adjust for yearly changes in the sampling Time series model building design, we applied the weights provided by the AHRQ Using the CCFs and clinical judgment to determine the HCUP for the NIS (9). Our institutional review board determined that this project was not human subjects appropriate leading/lagging relationship association research. All analyses were performed using R, version between the infl uenza and COPD series, we formulated 2.14.1 (R Foundation for Statistical Computing). four time series regression models with autocorrelated We fi rst identifi ed all hospitalizations during the errors. Th e errors were described using seasonal autore- period from January 1998 through July 2010 during gressive integrated moving average (ARIMA) models. In which a primary diagnosis of COPD was received. For each regression model, a COPD incidence series served case ascertainment, we used the International Classifi - as the response series and infl uenza activity served as th Revision, Clinical Modifi cation cation of Diseases, 9 the explanatory series. To better meet the assumption of (ICD-9-CM) codes 491.x (chronic bronchitis), 492.x stationarity, all of the series were log transformed, and (emphysema), 496 (chronic airway obstruction), and a fi rst-order seasonal diff erence was then applied to the 493.2x (chronic obstructive asthma). We then aggregated log-transformed series. all cases of COPD by month, based on the month that the Our time series regression models are summarized patient was admitted to the hospital. We also identifi ed in Table 1. Th e concurrent relationship between COPD Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 575 Table 1. Fitted time series regression models with the COPD incidence series as average components) may lead to incorrect inferential the response series and concurrent infl uenza activity as the explanatory series conclusions. Th e fi nal time series regression models, Coeffi cients Estimates SE p-value fi t using maximum likelihood, showed no evidence of lack of fi t, based on an inspection of the ACF and AR1 1.5013 0.1278 < 0.0001 PACF for the residuals. AR2 –0.5023 0.1264 < 0.0001 COPD MA1 –0.8827 0.0760 < 0.0001 Time Series Model Forecasting SMA1 –0.8996 0.2097 < 0.0001 To further confi rm the contemporaneous association FLU 0.0970 0.0108 < 0.0001 between each of the COPD series and infl uenza, we used AR1 0.9879 0.0197 < 0.0001 the novel H1N1 pandemic as a natural experiment. In COPD Requiring MA1 –0.5629 0.0969 < 0.0001 particular, we applied the existing time series models, fi t Mechanical using 1998–2008 data, to forecast monthly COPD admis- SMA1 –0.7494 0.1116 < 0.0001 Ventilation sions during 2009 and the fi rst half of 2010. Note that FLU 0.0965 0.0118 < 0.0001 this period includes the fall pandemic, which peaked in AR1 1.4741 0.1894 < 0.0001 October, as well as the months preceding and following. AR2 –0.4950 0.1754 0.0037 Specifi cally, we investigated whether the inclusion of exter- AECOPD MA1 –0.7648 0.1400 < 0.0001 nal infl uenza information would improve the forecasting SMA1 –0.8153 0.1593 < 0.0001 accuracy of COPD admissions during the early H1N1 out- FLU 0.1007 0.0120 < 0.0001 break period. A one-step-ahead forecasting scheme was implemented in our experiment based on the models with AR1 0.9756 0.0324 < 0.0001 and without infl uenza. Mean squared prediction errors AECOPD Requiring MA1 –0.4901 0.1030 < 0.0001 Mechanical (MSPEs) were employed to compare the forecasting per- SMA1 –0.7306 0.1054 < 0.0001 Ventilation formances of the two diff erent time series models. FLU 0.1026 0.0138 < 0.0001 AR1: Autoregressive component of order 1. Measure of Attributable Risk of Infl uenza for AR2: Autoregressive component of order 2. COPD Admissions MA1: Moving average component of order 1. SMA1: Seasonal moving average component of order 1 with a periodicity of 12. To further quantify the burden of infl uenza activity on FLU: Infl uenza activity. COPD incidence, we computed a measure of attribut- Note: The autoregressive and moving average components are included in the models able risk of COPD admissions due to infl uenza. First, we to account for autocorrelation in the residuals. The infl uenza coeffi cients represent the found the peak infl uenza month during each 12-month instantaneous associations between the infl uenza series and the COPD series. The instan- taneous associations were confi rmed through the cross-correlation functions. period from July of one year to June of the following year. We then calculated the excess risk of COPD admis- sions related to infl uenza in the peak infl uenza month incidence and infl uenza activity is indicated by the of each year by computing the diff erence between the CCFs for the four prewhitened series (Supplemental average rate of admissions for COPD during all twelve Figure S1). Th e CCFs exhibit strong statistically signifi - months, representing the overall risk, and the average cant peaks at lag zero, indicating that there is an instan- rate of admissions for COPD during the eleven non- taneous correlation between time of COPD admission peak months. Th e attributable risk for the year was then and infl uenza activity in the population, rather than defi ned as a ratio of the excess risk to the overall risk. a leading/lagging association between the two series. Th e fi nal attributable risk measure was based on the Th is is consistent with clinical judgment, which dictates average of the yearly ratios. that any elevation in the risk of COPD due to infl uenza Th e attributable risk measure refl ects the proportion would be expected to occur within the same month. All of the overall incidence of COPD hospitalizations that of the models feature an autoregressive component of could be potentially eliminated if infl uenza activity dur- order 1 (AR1), a moving average component of order ing the peak month could be reduced to the baseline level 1 (MA1), and a seasonal moving average component corresponding to the average over the eleven non-peak of order 1 with a periodicity of 12 (SMA1). An autore- months. For example, a measure of 0.03 implies that 3% gressive component of order 2 (AR2) is also contained of annual COPD hospitalizations could be conceivably in the models for COPD and AECOPD. Th ese compo- prevented if infl uenza during the annual peak month nents were suggested by the autocorrelation function could be held to the baseline annual level. (ACF) and the partial autocorrelation function (PACF) for the residuals from a simple linear regression model Results fi t to the response and explanatory series using ordi- nary least squares. By modeling the temporal patterns Time series modeling in the residuals, we reduce the standard errors and Using the CCF based on the prewhitened series, we found improve the sensitivity of the resulting inferential pro- signifi cant contemporaneous correlations between each cedures. Failure to account for the autocorrelation in of the COPD incidence series and infl uenza activity the residuals (i.e. with the autoregressive and moving (Online Supplemental Figure S1). Our time series www.copdjournal.com 576 Gerke et al. regression models, based on these contemporaneous cor- Forecasting COPD admissions Both the actual COPD series and the fitted COPD relations, indicate strong, signifi cant associations between infl uenza and COPD, infl uenza and COPD requiring series based on the 1998–2008 data, along with the mechanical ventilation, infl uenza and AECOPD, and two predicted COPD series with and without external infl uenza and AECOPD requiring mechanical ventilation inf luenza information in the 2009 inf luenza pandemic, (all p-values < 0.0001) (Table 1). All other components are displayed in the upper panel of Figure 1. The cor- incorporated into the fi nal models were signifi cant. Th is responding influenza incidence series is shown in the is likely due to additional autocorrelations and seasonal lower panel of Figure 1. The inclusion of the external variations that were not fully explained by the infl uenza influenza information in the model greatly improves series. In all of our models, concurrent infl uenza activity the forecasting performance of COPD admissions, signifi cantly improves the prediction of COPD hospital- especially in tracking the COPD admissions peak cor- ization incidence. When the infl uenza series is dropped responding to the unusually early influenza outbreak from any model, we observed a considerable increase in in September and October 2009, and the subsequent the value of the Akaike information criterion, indicating COPD admissions pattern after the outbreak . Figure 2 the importance of infl uenza in predicting admissions for shows the more detailed forecasting results regarding COPD. (For instance, the AIC for the overall COPD model the early outbreak. As illustrated, prediction without with the infl uenza series is –279.06; the AIC for the model the external influenza information fails to detect the without the infl uenza series is –220.39. A diff erence of two early outbreak and continues to incorrectly track is viewed as meaningful.) Th e overall COPD series, along COPD admissions, since the forecasts exploit the sea- with the fi tted series based on the fi rst model described sonal memory where the peak usually happens in a in Table 1, is displayed in Figure 1. Note that the model typical winter month. A 29.9% reduction in MSPE was provides highly accurate fi tted values of COPD incidence accomplished when our forecasting models used the during the study period. Similarly, at a regional level, we external influenza information. Similar patterns were observed a strong concurrent relationship between COPD also found for the AECOPD series where we observed series and infl uenza within each of the four diff erent cen- a 31.3% reduction in MSPE. Detailed forecasting sus regions (results not shown). results regarding AECOPD are shown in Figure 3. Figure 1. COPD admissions (upper panel) and infl uenza admissions (lower panel) by month from January 1998 to July 2010. In the upper panel, prior to 2009, the red series represents the fi tted values based on the time series model with concurrent infl uenza activity as an explanatory variable. After 2009, the dotted red series represents forecasts of COPD admissions with infl uenza; the dotted blue series represents forecasts of COPD admissions without infl uenza. Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 577 Figure 2. Time series forecast for the COPD admissions during July 2009 through June 2010. In the upper panel, the black series represents the actual COPD series; the dotted red series represents forecasts of COPD admissions with infl uenza and the dotted blue series represents forecasts of COPD admissions without infl uenza. The corresponding infl uenza series is shown in the lower panel. Importantly, the last 6 months of 2009 include the second wave of the 2009 infl uenza pandemic. Note: The right vertical axis represents monthly COPD admissions in terms of the percentage of peak monthly COPD admissions during the forecasting period. Thus, the peak month corresponds to 100%. For example, in December 2009, the forecasting error with infl uenza is roughly 1%, and the error without infl uenza is approximately 15% (where the percentage is relative to peak admissions during the forecasting period). Attributable risk of infl uenza for COPD admissions 65 years of age, the association is strongest for those Th e national COPD series were then partitioned into who require mechanical ventilation during the hospi- two age groups: under 65 and 65 and over. To character- talization. Our results also suggest that knowledge of ize the national impact of infl uenza activity on COPD concurrent infl uenza activity in the population can be incidence, we calculated our attributable risk measure used to substantially improve prediction of admissions for each of the four COPD incidence series stratifi ed by due to COPD. Because the peak in pandemic infl uenza age group (Table 2). Note that the attributable risk mea- was not in a typical winter month, the natural experi- sure of infl uenza activity on COPD incidence increases ment provided by this outbreak provides further evi- with age. Th e risk is also higher for patients admitted dence that infl uenza is a signifi cant driver of COPD for AECOPD who are 65 years and older and elderly hospitalizations. Improvements in infl uenza surveil- patients that require mechanical ventilation. lance, prevention, and treatment could provide signifi - cant opportunities to decrease the national burden of hospitalizations of patients with COPD. Discussion Our fi ndings further indicate that infl uenza surveil- Our results clearly show that the incidence of hospital- lance is important in anticipating the need for mechani- izations for COPD is signifi cantly associated with infl u- cal ventilation in patients with COPD. Anticipating high, enza activity. Based on our attributable risk measure, disproportionate, or cyclical increases in hospitalizations this association is most prominent for patients admitted and mechanical ventilation may allow hospitals to more specifi cally for AECOPD, and for patients with COPD effi ciently use equipment (ventilators, bilevel positive who are over 65 years of age. Among patients over airway pressure machines), staffi ng (specialized nursing, www.copdjournal.com 578 Gerke et al. Figure 3. Time series forecast for AECOPD admissions during July 2009 through June 2010. In the upper panel, the black series represents the actual AECOPD series; the dotted red series represents forecasts of AECOPD admissions with infl uenza and the dotted blue series represents forecasts of AECOPD admissions without infl uenza. The corresponding infl uenza series is shown in the lower panel. Importantly, the last 6 months of 2009 include the second wave of the 2009 infl uenza pandemic. Note: The right vertical axis represents monthly AECOPD admissions in terms of the percentage of peak monthly AECOPD admissions during the forecasting period. Thus, the peak month corresponds to 100%. For example, in December 2009, the forecasting error with infl uenza is roughly 2–3%, and the error without infl uenza is approximately 15% (where the percentage is relative to peak admissions during the forecasting period). With more local and timely infl uenza activity reports, Table 2. Attributable risk measures by age group for the four COPD series physicians can better target interventions for patients AECOPD with COPD prior to hospitalization. For example, COPD Requiring Requiring patients with COPD exacerbations could potentially Mechanical Mechanical COPD Ventilation AECOPD Ventilation benefi t from early antiviral or anti-infl ammatory treat- Under 65 years 0.0374 0.0363 0.0391 0.0375 ment strategies. Oseltamivir is currently recommended for treatment of infl uenza in hospitalized patients (15). 65 years and over 0.0419 0.0489 0.0423 0.0503 Th e drug not only reduces viral load, but also may have Note: The measure refl ects the proportion of the overall incidence of COPD hospitalizations that could be potentially eliminated if infl uenza activity during the peak month could be an anti-infl ammatory eff ect that could reduce sever- reduced to the baseline level corresponding to the average over the 11 non-peak months. ity of an exacerbation (16). However, to be eff ective, it must be administered promptly within the fi rst 48 respiratory therapists), and ward census (particularly hours of symptoms (15). Th us, more timely informa- with respect to intensive care units) based on infl uenza tion regarding infl uenza activity may help physicians activity. At this time in the U.S., the most widely avail- target and aggressively diagnose and treat patients able surveillance data is at the state level, which can often with COPD during the infl uenza season. In addition, be 1–2-weeks-old when it becomes available. However, randomized trials have shown that prior use of inhaled novel surveillance methods using internet search queries corticosteroids reduces hospitalizations in patients with through other forms of social media can also provide frequent exacerbations, and similar fi ndings have been timely estimates, and perhaps, even forecasts of infl uenza shown for long-acting bronchodilators (17–21). Over- activity (11–14). all, the impact of infl uenza on outcomes in patients with Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 579 COPD warrants further research to defi ne the exact population level with a large sample size of hospitalized pathological mechanisms in infl uenza-related fl ares in patients across the nation. Further, our study exploits order to develop novel therapies that may alone, or in the benefi ts of using a time series modeling framework combination, reduce the severity of COPD hospitaliza- and controls for spurious correlation induced by com- tions during the infl uenza season. mon seasonal patterns. Th is approach allows us to see Seasonal infl uenza vaccination is also eff ective the strong correlation in monthly variations of infl uenza in preventing infl uenza and appears to decrease the and COPD admissions over a period of 11 years. Fur- frequency of COPD exacerbations (22–25). However, thermore, the peak of the H1N1 infl uenza pandemic in despite clear guidelines for routine infl uenza vaccina- 2009 allowed us to perform a natural experiment to con- tion in patients with COPD, vaccination rates in the fi rm our associations and to test our prediction models, United States continue to be below the target rates as the peak incidence was in October, rather than in the of both the World Health Organization and the U.S. typical winter months. Public Health Service “Healthy People 2010” initiative In conclusion, our study found an association for all age and risk groups (26). Our results indicate between hospitalization for COPD and infl uenza activ- that there is still a signifi cant burden of COPD hos- ity in the United States on a national level. Th e associa- pitalizations related to infl uenza. Our attributable risk tion is pronounced among older patients, particularly measure indicates that if infl uenza activity in the peak in those needing mechanical ventilation, and patients month of the year alone could be decreased to the base- with acute exacerbations. Our results show that, line level that occurs during the rest of the year, then despite guidelines for routine vaccination in patients approximately 3–5% of COPD hospitalizations per with COPD, infl uenza continues to have a signifi cant year (approximately 18,000–30,000) could potentially infl uence on patient outcomes and healthcare use. Fur- be avoided. Th us, our results provide further evidence ther research is needed to prevent and treat infl uenza for the importance of vaccinating patients with COPD in patients with COPD. Finally, the use of improved against infl uenza. surveillance and development of novel surveillance Our study has several limitations. First, we use measures for infl uenza may help prevent, as well as administrative data rather than clinical or microbiologic forecast, hospital admissions. data for case ascertainment. ICD-9-CM codes have a reasonable sensitivity, specifi city, and positive predic- Acknowledgements tive value for detecting infl uenza (27,28). For COPD, the sensitivity of the ICD-9-CM code may be less strong Th e authors would like to acknowledge Alejandro than specifi city, and therefore, we may have missed Comellas, M.D. for his review and suggestions in the some cases using our methods (29). Second, other preparation of this manuscript. respiratory viral pathogens that we did not analyze may co-circulate during winter months, possibly contrib- Declaration of Interest Statement uting to COPD incidence. Th ird, our study is ecologi- cal. We used the aggregate incidence for each disease Th e authors report no confl icts of interest. Th is work and did not study associations at the individual level; was supported in part by the National Institutes of instead, we focused on infl uenza as an “environmen- Health, Grant 1KL2RR024980: Institute for Clinical and tal” risk factor. Although it would be ideal to have data Translational Science, University of Iowa (AKG) and by a showing that individual people actually had infl uenza National Institutes of Health Career Investigator Award immediately prior to their COPD hospitalization, we (Research Grant K01 AI75089) (PMP). Th e funding cannot directly infer this from hospital discharge data. sources did not have involvement in the study design, Fourth, we do not consider temperature or humidity in data analysis, writing, or submission of the manuscript. our models. However, our results show strong associa- tions between infl uenza and COPD admissions across References diff erent geographic regions in the United States. Last, because we do not know the percentage of patients with 1. Hurd S. Th e impact of COPD on lung health worldwide: epidemiology and incidence. Chest 2000; 117:1S–4S. COPD and infl uenza who do not get hospitalized, we 2. Toy EL, Gallagher KF, Stanley EL, Swensen AR, Duh MS. cannot calculate a traditional measure of attributable Th e economic impact of exacerbations of chronic obstructive risk. However, we did calculate a modifi ed risk value of pulmonary disease and exacerbation defi nition: a review. COPD hospitalizations for the peak month of infl uenza COPD 2010; 7:214–228. each year as compared to the risk during the non-peak 3. Halpern MT, Stanford RH, Borker R. Th e burden of COPD in the U.S.A.: results from the confronting COPD survey. Respir months. Th is is a conservative estimate, as it does not Med 2003; 97 Suppl C:S81–89. take into account the potential for eliminating infl uenza 4. Wedzicha JA, Seemungal TA. COPD exacerbations: defi ning altogether, or even lessening the incidence over the their cause and prevention. Lancet 2007; 370:786–796. eleven non-peak months. 5. Celli BR, MacNee W. Standards for the diagnosis and treatment Despite these limitations, we are able to explore the of patients with COPD: a summary of the ATS/ERS position paper. Eur Respir J 2004; 23:932–946. impact of infl uenza on COPD hospitalizations on a www.copdjournal.com 580 Gerke et al. 6. MacIntyre N, Huang YC. Acute exacerbations and respiratory inhaled tiotropium in chronic obstructive pulmonary disease. failure in chronic obstructive pulmonary disease. Proc Am Eur Respir J 2002; 19:217–224. Th orac Soc 2008; 5:530–535. 19. Paggiaro PL, Vagaggini B, Di Franco A, Zingoni M, Fano M, 7. Miravitlles M, Ferrer M, Pont A, Zalacain R, Alvarez-Sala JL, Biraghi M. Effi cacy of nebulized fl unisolide combined with Masa F, et al. Eff ect of exacerbations on quality of life in salbutamol and ipratropium bromide in stable patients with patients with chronic obstructive pulmonary disease: a 2 year moderate-to-severe chronic obstructive pulmonary disease. follow up study. Th orax 2004; 59:387–395. Respiration 2006; 73:603–609. 8. Moineddin R, Nie JX, Domb G, Leong AM, Upshur RE. 20. Calverley PM, Boonsawat W, Cseke Z, Zhong N, Peterson S, Seasonality of primary care utilization for respiratory diseases Olsson H. Maintenance therapy with budesonide and in Ontario: a time-series analysis. BMC Health Serv Res 2008; formoterol in chronic obstructive pulmonary disease. Eur 8:160. Respir J 2003; 22:912–919. 9. HCUP Nationwide Inpatient Sample (NIS). Healthcare Cost 21. Brusasco V, Hodder R, Miravitlles M, Korducki L, Towse L, and Utilization Project (HCUP) [Internet]. Rockville, MD: Kesten S. Health outcomes following treatment for six Agency for Healthcare Research and Quality, 1998–2010 months with once daily tiotropium compared with twice [updated 2012 Sep 10; cited 2013 Jan 14]; Available from: daily salmeterol in patients with COPD. Th orax 2003; www.hcup-us.ahrq.gov/nisoverview.jsp 58:399–404. 10. Cryer J. Time Series Analysis with Applications in R. New 22. Menon B, Gurnani M, Aggarwal B. Comparison of outpatient York: Springer, 2008. visits and hospitalisations, in patients with chronic obstructive 11. Polgreen PM, Chen Y, Pennock DM, Nelson FD. Using pulmonary disease, before and after infl uenza vaccination. Int internet searches for infl uenza surveillance. Clin Infect Dis J Clin Pract 2008; 62:593–598. 2008; 47:1443–1448. 23. Poole PJ, Chacko E, Wood-Baker RW, Cates CJ. Infl uenza 12. Signorini A, Segre AM, Polgreen PM. Th e use of Twitter to vaccine for patients with chronic obstructive pulmonary track levels of disease activity and public concern in the U.S. disease. Cochrane Database Syst Rev 2006 Jan;1:CD002733. during the infl uenza A H1N1 pandemic. PLoS One 2011; 24. S chembr i S , Morant S , Winter JH, M acD onald TM. Inf luenz a 6:e19467. but not pneumococcal vaccination protects against all- 13. Ginsberg J, Mohebbi MH, Patel RS, Brammer L, Smolinski cause mortality in patients with COPD. Thorax 2009; MS, Brilliant L. Detecting infl uenza epidemics using search 64:567–572. engine query data. Nature 2009; 457:1012–1014. 25. Wongsurakiat P, Maranetra KN, Wasi C, Kositanont U, 14. Corley CD, Cook DJ, Mikler AR, Singh KP. Using web and Dejsomritrutai W, Charoenratanakul S. Acute respiratory social media for infl uenza surveillance. Adv Exp Med Biol illness in patients with COPD and the eff ectiveness of 2010; 680:559–564. infl uenza vaccination: a randomized controlled study. Chest 15. Mandell LA, Wunderink RG, Anzueto A, Bartlett JG, 2004; 125:2011–2020. Campbell GD, Dean NC, et al. Infectious Diseases Society 26. Monto AS. Seasonal infl uenza and vaccination coverage. of America/American Th oracic Society consensus guidelines Vaccine 2010; 28 Suppl 4:D33–44. on the management of community-acquired pneumonia in 27. Ginde AA, Tsai CL, Blanc PG, Camargo CA, Jr. Positive adults. Clin Infect Dis 2007; 44 Suppl 2:S27–72. predictive value of ICD-9-CM codes to detect acute 16. Wong ZX, Jones JE, Anderson GP, Gualano RC. Oseltamivir exacerbation of COPD in the emergency department. Jt Comm treatment of mice before or after mild infl uenza infection J Qual Patient Saf 2008; 34:678–680. reduced cellular and cytokine infl ammation in the lung. 28. Marsden-Haug N, Foster VB, Gould PL, Elbert E, Wang H, Infl uenza Other Respi Viruses 2011; 5(5):343–350. Pavlin JA . Code-based syndromic surveillance for infl uenzalike 17. Vincken W, van Noord JA, Greefhorst AP, Bantje TA, Kesten S, illness by International Classifi cation of Diseases, Ninth Korducki L, et al. Improved health outcomes in patients with Revision. Emerg Infect Dis 2007;13:207–216. COPD during 1 yr’s treatment with tiotropium. Eur Respir J 29. Stein BD, Bautista A, Schumock GT, Lee TA, Charbeneau JT, 2002; 19:209–216. Lauderdale DS, et al. Th e validity of ICD-9-CM diagnosis codes 18. Casaburi R, Mahler DA, Jones PW, Wanner A, San PG, for identifying patients hospitalized for COPD exacerbations. ZuWallack RL, et al. A long-term evaluation of once-daily Chest 2012; 141(1):87–93. A supplemental fi gure is available in the online version of this article. Copyright © 2013 Informa Healthcare USA, Inc http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png COPD Journal of Chronic Obstructive Pulmonary Disease Taylor & Francis

Predicting Chronic Obstructive Pulmonary Disease Hospitalizations Based on Concurrent Influenza Activity

Download PDF
8 pages

Loading next page...
 
/lp/taylor-francis/predicting-chronic-obstructive-pulmonary-disease-hospitalizations-IE6HnFMdW4

References (31)

Publisher
Taylor & Francis
Copyright
© 2013 Informa Healthcare USA, Inc.
ISSN
1541-2563
eISSN
1541-2555
DOI
10.3109/15412555.2013.777400
pmid
23819753
Publisher site
See Article on Publisher Site

Abstract

2 University of Iowa Department of Biostatistics, Iowa City, IA, USA Although infl uenza has been associated with chronic obstructive pulmonary 3 Harvard School of Public Health: 651 Huntington disease (COPD) exacerbations, it is not clear the extent to which this association Avenue, Boston, MA, USA affects healthcare use in the United States. The fi rst goal of this project was to determine to what extent the incidence of COPD hospitalizations is associated with seasonal infl uenza. Second, as a natural experiment, we used infl uenza activity to help predict COPD admissions during the 2009 H1N1 infl uenza pandemic. To do this, we identifi ed all hospitalizations between 1998 and 2010 in the Nationwide Inpatient Sample from the Healthcare Cost and Utilization Project (HCUP) during which a primary diagnosis of COPD was recorded. Separately, we identifi ed all hospitalizations during which a diagnosis of infl uenza was recorded. We formulated time series regression models to investigate the association of monthly COPD admissions with infl uenza incidence. Finally, we applied these models, fi t using 1998–2008 data, to forecast monthly COPD admissions during the 2009 pandemic. Based on time series regression models, a strong, signifi cant association exists between concurrent infl uenza activity and incidence of COPD hospitalizations (p-value < 0.0001). The association is especially strong among older patients requiring mechanical ventilation. Use of infl uenza data to predict COPD admissions during the 2009 H1N1 pandemic reduced the mean-squared prediction error by 29.9%. We conclude that infl uenza activity is signifi cantly associated with COPD hospitalizations in the United States and infl uenza activity can be exploited to more accurately forecast COPD admissions. Our results suggest that improvements in infl uenza surveillance, prevention, and treatment may decrease hospitalizations of patients diagnosed with COPD. Introduction Chronic obstructive pulmonary disease (COPD) is a major cause of morbid- f morbid- bi id- d is one of t o of t f th he e ity and mortality in the United States (1). In addition, COPD is one of the major or c r com o ompo po- o- - most expensive chronic diseases, and hospitalizations are a major compo- cute exac xac ac ce e erb rb ba atio t tions ion nent of direct medical costs attributable to COPD (2, 3). Acute exacerbations matory ry e e ep ep piso i iso isod de de es sst s tr t ig- ig g g g- of COPD (AECOPD) are thought to be acute infl ammatory episodes trig- Keywords: Chronic airfl ow obstruction, human infl uenza, epidemiology, hospitalization, disease pathogens ge ens, v , i ir r ru u us use ses s,, and a an and nd gered by environmental factors, including bacterial pathogens, viruses, and exacerbation, chronic obstructive pulmonary disease, en infl uenz uen nz nza a a and AEC an and nd dA AEC EC CO OPD O OPD D ambient air pollution (4, 5). Th e association between infl uenza and AECOPD COPD exacerbation. g epithe he el l lial c ial ia ce el lls and r lsa s nd d r re e e es sult sult ta an ant t is thought to be mediated via destruction of lung epithelial cells and resultant increa as s se e ed dp dp d prr ro o oduc du d duc u tion of m io on o mu m mu u uc cu u us s s, ,, infl ammation leading to lung tissue damage, increased production of mucus, Correspondence to: Philip M. Polgreen, MD, n. ThThe e e e e ex x xac a t d t dis di i ea as se e me em e ec chanisms ch ch nism sm m vasoconstriction, and bronchoconstriction. Th e exact disease mechanisms University of Iowa Department of Internal Medicine: 200 Hawkins Dr., Iowa City, Iowa 52242, er, episo iso od d de es of r of f r f r re e espira s spir pira ra at tor ry yyf f f failur a lur urea e e and nd d d are not completely understood. However, episodes of respiratory failure and phone: 319-356-1616, fax: 319-353-6406, sociate ed ed d w d w d w wi it i it th h ac h ac ac acc ce c celera era ra ate ted de d d de decline of lung ine f lu lu un n mechanical ventilation have been associated with accelerated decline of lung email; [email protected] n patien nt ts ts s diag d dia ag gn gn no nose e ed d w dw w wi it th h h C h C C CO OPD (6, 7). PD PD D D(6, 7 7 7) function and poor quality of life in patients diagnosed with COPD (6, 7). 573 573 574 Gerke et al. Epidemiologically, the seasonality of infl uenza infec- a subgroup of patients admitted specifi cally for acute tions and COPD exacerbations is similar, with both COPD exacerbations (ICD-9-CM code: 491.21). peaking during winter months from December to Feb- For both of these groups, we then identifi ed patients ruary (8). Th us, although infl uenza has been associated who required mechanical ventilation using the ICD-9- with COPD exacerbations, it is less clear how important CM codes 93.90 (non-invasive mechanical ventilation) these patterns are regarding healthcare use. Under- and/or 96.70–96.72 (continuous invasive mechanical standing the extent of the association between infl uenza ventilation). Separately, to build our infl uenza series, and COPD hospitalizations could lead to more focused we identifi ed all hospitalizations from January 1998 and cost-eff ective eff orts to prevent hospitalizations and through July 2010 during which a primary or secondary develop novel interventions. Predicting and quantifying diagnosis of infl uenza was received. For infl uenza case the burden of COPD hospitalizations using infl uenza- ascertainment, we used the ICD-9-CM codes 487.00 related data streams may also allow hospitals to plan for (infl uenza with pneumonia), 487.10 (infl uenza with increases in hospital resources (e.g. pharmaceuticals, other respiratory manifestations), and 487.80 (infl uenza ventilators, and staffi ng of critical care units). with other manifestations). We then compiled monthly Th e purpose of this study is to use time series meth- totals of cases of infl uenza in the same manner as COPD ods to determine whether, and to what extent, the inci- based on the month that the patient was admitted. Using dence of COPD hospitalizations is associated with the this same approach, we compiled COPD and infl uenza seasonal variation in the incidence of infl uenza. Because series from the four census regions (Northeast, Midwest, both infl uenza and COPD may be caused by some other South, and West) to confi rm the results at the national seasonal and winter-related factor, we use the unusu- level. ally early peak of the 2009 H1N1 infl uenza pandemic in Time series correlation analysis October as a natural experiment to determine if infl u- To investigate the association of COPD with infl uenza, enza and COPD admissions are associated outside of the traditional infl uenza season typically comprised of we computed cross-correlation functions (CCF) for winter months (December through February). Finally, the infl uenza series and each of our four COPD series: we provide estimates of a measure of attributable risk COPD, AECOPD, COPD with mechanical ventilation, that further characterize the relationship between and AECOPD with mechanical ventilation. Th e CCF COPD and infl uenza. indicates the temporal correlations between two time series: specifi cally, a series at time t and another series at time t+m, where m is referred to as the lag. Because cross- Methods correlations between time series can be spurious due to Data source the eff ects of common temporal patterns, we employed All data were extracted from the Nationwide Inpatient a prewhitening process (10). In our application, com- Sample (NIS), the largest all-payer database of national mon yearly cycles are present in both the COPD series discharges in the U.S. Th e database is maintained as and the infl uenza series, since both are elevated during part of the Healthcare Cost and Utilization Project the winter months. Th e prewhitening process allows us (HCUP) by the Agency for Healthcare Research and to detect correlations based on prominent local peaks or Quality (AHRQ), and contains data from a 20% strati- troughs in two time series that are temporally aligned, fi ed sample of nonfederal acute care hospitals (9). Th is as opposed to coincidental correlations based on shared sample includes academic medical centers, community seasonal patterns. Th e former are representative of a hospitals, general hospitals, and specialty hospitals. legitimate association, whereas the latter are merely due It excludes long-term care facilities and rehabilitation to common cyclic behavior. hospitals. To adjust for yearly changes in the sampling Time series model building design, we applied the weights provided by the AHRQ Using the CCFs and clinical judgment to determine the HCUP for the NIS (9). Our institutional review board determined that this project was not human subjects appropriate leading/lagging relationship association research. All analyses were performed using R, version between the infl uenza and COPD series, we formulated 2.14.1 (R Foundation for Statistical Computing). four time series regression models with autocorrelated We fi rst identifi ed all hospitalizations during the errors. Th e errors were described using seasonal autore- period from January 1998 through July 2010 during gressive integrated moving average (ARIMA) models. In which a primary diagnosis of COPD was received. For each regression model, a COPD incidence series served case ascertainment, we used the International Classifi - as the response series and infl uenza activity served as th Revision, Clinical Modifi cation cation of Diseases, 9 the explanatory series. To better meet the assumption of (ICD-9-CM) codes 491.x (chronic bronchitis), 492.x stationarity, all of the series were log transformed, and (emphysema), 496 (chronic airway obstruction), and a fi rst-order seasonal diff erence was then applied to the 493.2x (chronic obstructive asthma). We then aggregated log-transformed series. all cases of COPD by month, based on the month that the Our time series regression models are summarized patient was admitted to the hospital. We also identifi ed in Table 1. Th e concurrent relationship between COPD Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 575 Table 1. Fitted time series regression models with the COPD incidence series as average components) may lead to incorrect inferential the response series and concurrent infl uenza activity as the explanatory series conclusions. Th e fi nal time series regression models, Coeffi cients Estimates SE p-value fi t using maximum likelihood, showed no evidence of lack of fi t, based on an inspection of the ACF and AR1 1.5013 0.1278 < 0.0001 PACF for the residuals. AR2 –0.5023 0.1264 < 0.0001 COPD MA1 –0.8827 0.0760 < 0.0001 Time Series Model Forecasting SMA1 –0.8996 0.2097 < 0.0001 To further confi rm the contemporaneous association FLU 0.0970 0.0108 < 0.0001 between each of the COPD series and infl uenza, we used AR1 0.9879 0.0197 < 0.0001 the novel H1N1 pandemic as a natural experiment. In COPD Requiring MA1 –0.5629 0.0969 < 0.0001 particular, we applied the existing time series models, fi t Mechanical using 1998–2008 data, to forecast monthly COPD admis- SMA1 –0.7494 0.1116 < 0.0001 Ventilation sions during 2009 and the fi rst half of 2010. Note that FLU 0.0965 0.0118 < 0.0001 this period includes the fall pandemic, which peaked in AR1 1.4741 0.1894 < 0.0001 October, as well as the months preceding and following. AR2 –0.4950 0.1754 0.0037 Specifi cally, we investigated whether the inclusion of exter- AECOPD MA1 –0.7648 0.1400 < 0.0001 nal infl uenza information would improve the forecasting SMA1 –0.8153 0.1593 < 0.0001 accuracy of COPD admissions during the early H1N1 out- FLU 0.1007 0.0120 < 0.0001 break period. A one-step-ahead forecasting scheme was implemented in our experiment based on the models with AR1 0.9756 0.0324 < 0.0001 and without infl uenza. Mean squared prediction errors AECOPD Requiring MA1 –0.4901 0.1030 < 0.0001 Mechanical (MSPEs) were employed to compare the forecasting per- SMA1 –0.7306 0.1054 < 0.0001 Ventilation formances of the two diff erent time series models. FLU 0.1026 0.0138 < 0.0001 AR1: Autoregressive component of order 1. Measure of Attributable Risk of Infl uenza for AR2: Autoregressive component of order 2. COPD Admissions MA1: Moving average component of order 1. SMA1: Seasonal moving average component of order 1 with a periodicity of 12. To further quantify the burden of infl uenza activity on FLU: Infl uenza activity. COPD incidence, we computed a measure of attribut- Note: The autoregressive and moving average components are included in the models able risk of COPD admissions due to infl uenza. First, we to account for autocorrelation in the residuals. The infl uenza coeffi cients represent the found the peak infl uenza month during each 12-month instantaneous associations between the infl uenza series and the COPD series. The instan- taneous associations were confi rmed through the cross-correlation functions. period from July of one year to June of the following year. We then calculated the excess risk of COPD admis- sions related to infl uenza in the peak infl uenza month incidence and infl uenza activity is indicated by the of each year by computing the diff erence between the CCFs for the four prewhitened series (Supplemental average rate of admissions for COPD during all twelve Figure S1). Th e CCFs exhibit strong statistically signifi - months, representing the overall risk, and the average cant peaks at lag zero, indicating that there is an instan- rate of admissions for COPD during the eleven non- taneous correlation between time of COPD admission peak months. Th e attributable risk for the year was then and infl uenza activity in the population, rather than defi ned as a ratio of the excess risk to the overall risk. a leading/lagging association between the two series. Th e fi nal attributable risk measure was based on the Th is is consistent with clinical judgment, which dictates average of the yearly ratios. that any elevation in the risk of COPD due to infl uenza Th e attributable risk measure refl ects the proportion would be expected to occur within the same month. All of the overall incidence of COPD hospitalizations that of the models feature an autoregressive component of could be potentially eliminated if infl uenza activity dur- order 1 (AR1), a moving average component of order ing the peak month could be reduced to the baseline level 1 (MA1), and a seasonal moving average component corresponding to the average over the eleven non-peak of order 1 with a periodicity of 12 (SMA1). An autore- months. For example, a measure of 0.03 implies that 3% gressive component of order 2 (AR2) is also contained of annual COPD hospitalizations could be conceivably in the models for COPD and AECOPD. Th ese compo- prevented if infl uenza during the annual peak month nents were suggested by the autocorrelation function could be held to the baseline annual level. (ACF) and the partial autocorrelation function (PACF) for the residuals from a simple linear regression model Results fi t to the response and explanatory series using ordi- nary least squares. By modeling the temporal patterns Time series modeling in the residuals, we reduce the standard errors and Using the CCF based on the prewhitened series, we found improve the sensitivity of the resulting inferential pro- signifi cant contemporaneous correlations between each cedures. Failure to account for the autocorrelation in of the COPD incidence series and infl uenza activity the residuals (i.e. with the autoregressive and moving (Online Supplemental Figure S1). Our time series www.copdjournal.com 576 Gerke et al. regression models, based on these contemporaneous cor- Forecasting COPD admissions Both the actual COPD series and the fitted COPD relations, indicate strong, signifi cant associations between infl uenza and COPD, infl uenza and COPD requiring series based on the 1998–2008 data, along with the mechanical ventilation, infl uenza and AECOPD, and two predicted COPD series with and without external infl uenza and AECOPD requiring mechanical ventilation inf luenza information in the 2009 inf luenza pandemic, (all p-values < 0.0001) (Table 1). All other components are displayed in the upper panel of Figure 1. The cor- incorporated into the fi nal models were signifi cant. Th is responding influenza incidence series is shown in the is likely due to additional autocorrelations and seasonal lower panel of Figure 1. The inclusion of the external variations that were not fully explained by the infl uenza influenza information in the model greatly improves series. In all of our models, concurrent infl uenza activity the forecasting performance of COPD admissions, signifi cantly improves the prediction of COPD hospital- especially in tracking the COPD admissions peak cor- ization incidence. When the infl uenza series is dropped responding to the unusually early influenza outbreak from any model, we observed a considerable increase in in September and October 2009, and the subsequent the value of the Akaike information criterion, indicating COPD admissions pattern after the outbreak . Figure 2 the importance of infl uenza in predicting admissions for shows the more detailed forecasting results regarding COPD. (For instance, the AIC for the overall COPD model the early outbreak. As illustrated, prediction without with the infl uenza series is –279.06; the AIC for the model the external influenza information fails to detect the without the infl uenza series is –220.39. A diff erence of two early outbreak and continues to incorrectly track is viewed as meaningful.) Th e overall COPD series, along COPD admissions, since the forecasts exploit the sea- with the fi tted series based on the fi rst model described sonal memory where the peak usually happens in a in Table 1, is displayed in Figure 1. Note that the model typical winter month. A 29.9% reduction in MSPE was provides highly accurate fi tted values of COPD incidence accomplished when our forecasting models used the during the study period. Similarly, at a regional level, we external influenza information. Similar patterns were observed a strong concurrent relationship between COPD also found for the AECOPD series where we observed series and infl uenza within each of the four diff erent cen- a 31.3% reduction in MSPE. Detailed forecasting sus regions (results not shown). results regarding AECOPD are shown in Figure 3. Figure 1. COPD admissions (upper panel) and infl uenza admissions (lower panel) by month from January 1998 to July 2010. In the upper panel, prior to 2009, the red series represents the fi tted values based on the time series model with concurrent infl uenza activity as an explanatory variable. After 2009, the dotted red series represents forecasts of COPD admissions with infl uenza; the dotted blue series represents forecasts of COPD admissions without infl uenza. Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 577 Figure 2. Time series forecast for the COPD admissions during July 2009 through June 2010. In the upper panel, the black series represents the actual COPD series; the dotted red series represents forecasts of COPD admissions with infl uenza and the dotted blue series represents forecasts of COPD admissions without infl uenza. The corresponding infl uenza series is shown in the lower panel. Importantly, the last 6 months of 2009 include the second wave of the 2009 infl uenza pandemic. Note: The right vertical axis represents monthly COPD admissions in terms of the percentage of peak monthly COPD admissions during the forecasting period. Thus, the peak month corresponds to 100%. For example, in December 2009, the forecasting error with infl uenza is roughly 1%, and the error without infl uenza is approximately 15% (where the percentage is relative to peak admissions during the forecasting period). Attributable risk of infl uenza for COPD admissions 65 years of age, the association is strongest for those Th e national COPD series were then partitioned into who require mechanical ventilation during the hospi- two age groups: under 65 and 65 and over. To character- talization. Our results also suggest that knowledge of ize the national impact of infl uenza activity on COPD concurrent infl uenza activity in the population can be incidence, we calculated our attributable risk measure used to substantially improve prediction of admissions for each of the four COPD incidence series stratifi ed by due to COPD. Because the peak in pandemic infl uenza age group (Table 2). Note that the attributable risk mea- was not in a typical winter month, the natural experi- sure of infl uenza activity on COPD incidence increases ment provided by this outbreak provides further evi- with age. Th e risk is also higher for patients admitted dence that infl uenza is a signifi cant driver of COPD for AECOPD who are 65 years and older and elderly hospitalizations. Improvements in infl uenza surveil- patients that require mechanical ventilation. lance, prevention, and treatment could provide signifi - cant opportunities to decrease the national burden of hospitalizations of patients with COPD. Discussion Our fi ndings further indicate that infl uenza surveil- Our results clearly show that the incidence of hospital- lance is important in anticipating the need for mechani- izations for COPD is signifi cantly associated with infl u- cal ventilation in patients with COPD. Anticipating high, enza activity. Based on our attributable risk measure, disproportionate, or cyclical increases in hospitalizations this association is most prominent for patients admitted and mechanical ventilation may allow hospitals to more specifi cally for AECOPD, and for patients with COPD effi ciently use equipment (ventilators, bilevel positive who are over 65 years of age. Among patients over airway pressure machines), staffi ng (specialized nursing, www.copdjournal.com 578 Gerke et al. Figure 3. Time series forecast for AECOPD admissions during July 2009 through June 2010. In the upper panel, the black series represents the actual AECOPD series; the dotted red series represents forecasts of AECOPD admissions with infl uenza and the dotted blue series represents forecasts of AECOPD admissions without infl uenza. The corresponding infl uenza series is shown in the lower panel. Importantly, the last 6 months of 2009 include the second wave of the 2009 infl uenza pandemic. Note: The right vertical axis represents monthly AECOPD admissions in terms of the percentage of peak monthly AECOPD admissions during the forecasting period. Thus, the peak month corresponds to 100%. For example, in December 2009, the forecasting error with infl uenza is roughly 2–3%, and the error without infl uenza is approximately 15% (where the percentage is relative to peak admissions during the forecasting period). With more local and timely infl uenza activity reports, Table 2. Attributable risk measures by age group for the four COPD series physicians can better target interventions for patients AECOPD with COPD prior to hospitalization. For example, COPD Requiring Requiring patients with COPD exacerbations could potentially Mechanical Mechanical COPD Ventilation AECOPD Ventilation benefi t from early antiviral or anti-infl ammatory treat- Under 65 years 0.0374 0.0363 0.0391 0.0375 ment strategies. Oseltamivir is currently recommended for treatment of infl uenza in hospitalized patients (15). 65 years and over 0.0419 0.0489 0.0423 0.0503 Th e drug not only reduces viral load, but also may have Note: The measure refl ects the proportion of the overall incidence of COPD hospitalizations that could be potentially eliminated if infl uenza activity during the peak month could be an anti-infl ammatory eff ect that could reduce sever- reduced to the baseline level corresponding to the average over the 11 non-peak months. ity of an exacerbation (16). However, to be eff ective, it must be administered promptly within the fi rst 48 respiratory therapists), and ward census (particularly hours of symptoms (15). Th us, more timely informa- with respect to intensive care units) based on infl uenza tion regarding infl uenza activity may help physicians activity. At this time in the U.S., the most widely avail- target and aggressively diagnose and treat patients able surveillance data is at the state level, which can often with COPD during the infl uenza season. In addition, be 1–2-weeks-old when it becomes available. However, randomized trials have shown that prior use of inhaled novel surveillance methods using internet search queries corticosteroids reduces hospitalizations in patients with through other forms of social media can also provide frequent exacerbations, and similar fi ndings have been timely estimates, and perhaps, even forecasts of infl uenza shown for long-acting bronchodilators (17–21). Over- activity (11–14). all, the impact of infl uenza on outcomes in patients with Copyright © 2013 Informa Healthcare USA, Inc Infl uenza Activity Predicts COPD Hospitalizations 579 COPD warrants further research to defi ne the exact population level with a large sample size of hospitalized pathological mechanisms in infl uenza-related fl ares in patients across the nation. Further, our study exploits order to develop novel therapies that may alone, or in the benefi ts of using a time series modeling framework combination, reduce the severity of COPD hospitaliza- and controls for spurious correlation induced by com- tions during the infl uenza season. mon seasonal patterns. Th is approach allows us to see Seasonal infl uenza vaccination is also eff ective the strong correlation in monthly variations of infl uenza in preventing infl uenza and appears to decrease the and COPD admissions over a period of 11 years. Fur- frequency of COPD exacerbations (22–25). However, thermore, the peak of the H1N1 infl uenza pandemic in despite clear guidelines for routine infl uenza vaccina- 2009 allowed us to perform a natural experiment to con- tion in patients with COPD, vaccination rates in the fi rm our associations and to test our prediction models, United States continue to be below the target rates as the peak incidence was in October, rather than in the of both the World Health Organization and the U.S. typical winter months. Public Health Service “Healthy People 2010” initiative In conclusion, our study found an association for all age and risk groups (26). Our results indicate between hospitalization for COPD and infl uenza activ- that there is still a signifi cant burden of COPD hos- ity in the United States on a national level. Th e associa- pitalizations related to infl uenza. Our attributable risk tion is pronounced among older patients, particularly measure indicates that if infl uenza activity in the peak in those needing mechanical ventilation, and patients month of the year alone could be decreased to the base- with acute exacerbations. Our results show that, line level that occurs during the rest of the year, then despite guidelines for routine vaccination in patients approximately 3–5% of COPD hospitalizations per with COPD, infl uenza continues to have a signifi cant year (approximately 18,000–30,000) could potentially infl uence on patient outcomes and healthcare use. Fur- be avoided. Th us, our results provide further evidence ther research is needed to prevent and treat infl uenza for the importance of vaccinating patients with COPD in patients with COPD. Finally, the use of improved against infl uenza. surveillance and development of novel surveillance Our study has several limitations. First, we use measures for infl uenza may help prevent, as well as administrative data rather than clinical or microbiologic forecast, hospital admissions. data for case ascertainment. ICD-9-CM codes have a reasonable sensitivity, specifi city, and positive predic- Acknowledgements tive value for detecting infl uenza (27,28). For COPD, the sensitivity of the ICD-9-CM code may be less strong Th e authors would like to acknowledge Alejandro than specifi city, and therefore, we may have missed Comellas, M.D. for his review and suggestions in the some cases using our methods (29). Second, other preparation of this manuscript. respiratory viral pathogens that we did not analyze may co-circulate during winter months, possibly contrib- Declaration of Interest Statement uting to COPD incidence. Th ird, our study is ecologi- cal. We used the aggregate incidence for each disease Th e authors report no confl icts of interest. Th is work and did not study associations at the individual level; was supported in part by the National Institutes of instead, we focused on infl uenza as an “environmen- Health, Grant 1KL2RR024980: Institute for Clinical and tal” risk factor. Although it would be ideal to have data Translational Science, University of Iowa (AKG) and by a showing that individual people actually had infl uenza National Institutes of Health Career Investigator Award immediately prior to their COPD hospitalization, we (Research Grant K01 AI75089) (PMP). Th e funding cannot directly infer this from hospital discharge data. sources did not have involvement in the study design, Fourth, we do not consider temperature or humidity in data analysis, writing, or submission of the manuscript. our models. However, our results show strong associa- tions between infl uenza and COPD admissions across References diff erent geographic regions in the United States. Last, because we do not know the percentage of patients with 1. Hurd S. Th e impact of COPD on lung health worldwide: epidemiology and incidence. Chest 2000; 117:1S–4S. COPD and infl uenza who do not get hospitalized, we 2. Toy EL, Gallagher KF, Stanley EL, Swensen AR, Duh MS. cannot calculate a traditional measure of attributable Th e economic impact of exacerbations of chronic obstructive risk. However, we did calculate a modifi ed risk value of pulmonary disease and exacerbation defi nition: a review. COPD hospitalizations for the peak month of infl uenza COPD 2010; 7:214–228. each year as compared to the risk during the non-peak 3. Halpern MT, Stanford RH, Borker R. Th e burden of COPD in the U.S.A.: results from the confronting COPD survey. Respir months. Th is is a conservative estimate, as it does not Med 2003; 97 Suppl C:S81–89. take into account the potential for eliminating infl uenza 4. Wedzicha JA, Seemungal TA. COPD exacerbations: defi ning altogether, or even lessening the incidence over the their cause and prevention. Lancet 2007; 370:786–796. eleven non-peak months. 5. Celli BR, MacNee W. Standards for the diagnosis and treatment Despite these limitations, we are able to explore the of patients with COPD: a summary of the ATS/ERS position paper. Eur Respir J 2004; 23:932–946. impact of infl uenza on COPD hospitalizations on a www.copdjournal.com 580 Gerke et al. 6. MacIntyre N, Huang YC. Acute exacerbations and respiratory inhaled tiotropium in chronic obstructive pulmonary disease. failure in chronic obstructive pulmonary disease. Proc Am Eur Respir J 2002; 19:217–224. Th orac Soc 2008; 5:530–535. 19. Paggiaro PL, Vagaggini B, Di Franco A, Zingoni M, Fano M, 7. Miravitlles M, Ferrer M, Pont A, Zalacain R, Alvarez-Sala JL, Biraghi M. Effi cacy of nebulized fl unisolide combined with Masa F, et al. Eff ect of exacerbations on quality of life in salbutamol and ipratropium bromide in stable patients with patients with chronic obstructive pulmonary disease: a 2 year moderate-to-severe chronic obstructive pulmonary disease. follow up study. Th orax 2004; 59:387–395. Respiration 2006; 73:603–609. 8. Moineddin R, Nie JX, Domb G, Leong AM, Upshur RE. 20. Calverley PM, Boonsawat W, Cseke Z, Zhong N, Peterson S, Seasonality of primary care utilization for respiratory diseases Olsson H. Maintenance therapy with budesonide and in Ontario: a time-series analysis. BMC Health Serv Res 2008; formoterol in chronic obstructive pulmonary disease. Eur 8:160. Respir J 2003; 22:912–919. 9. HCUP Nationwide Inpatient Sample (NIS). Healthcare Cost 21. Brusasco V, Hodder R, Miravitlles M, Korducki L, Towse L, and Utilization Project (HCUP) [Internet]. Rockville, MD: Kesten S. Health outcomes following treatment for six Agency for Healthcare Research and Quality, 1998–2010 months with once daily tiotropium compared with twice [updated 2012 Sep 10; cited 2013 Jan 14]; Available from: daily salmeterol in patients with COPD. Th orax 2003; www.hcup-us.ahrq.gov/nisoverview.jsp 58:399–404. 10. Cryer J. Time Series Analysis with Applications in R. New 22. Menon B, Gurnani M, Aggarwal B. Comparison of outpatient York: Springer, 2008. visits and hospitalisations, in patients with chronic obstructive 11. Polgreen PM, Chen Y, Pennock DM, Nelson FD. Using pulmonary disease, before and after infl uenza vaccination. Int internet searches for infl uenza surveillance. Clin Infect Dis J Clin Pract 2008; 62:593–598. 2008; 47:1443–1448. 23. Poole PJ, Chacko E, Wood-Baker RW, Cates CJ. Infl uenza 12. Signorini A, Segre AM, Polgreen PM. Th e use of Twitter to vaccine for patients with chronic obstructive pulmonary track levels of disease activity and public concern in the U.S. disease. Cochrane Database Syst Rev 2006 Jan;1:CD002733. during the infl uenza A H1N1 pandemic. PLoS One 2011; 24. S chembr i S , Morant S , Winter JH, M acD onald TM. Inf luenz a 6:e19467. but not pneumococcal vaccination protects against all- 13. Ginsberg J, Mohebbi MH, Patel RS, Brammer L, Smolinski cause mortality in patients with COPD. Thorax 2009; MS, Brilliant L. Detecting infl uenza epidemics using search 64:567–572. engine query data. Nature 2009; 457:1012–1014. 25. Wongsurakiat P, Maranetra KN, Wasi C, Kositanont U, 14. Corley CD, Cook DJ, Mikler AR, Singh KP. Using web and Dejsomritrutai W, Charoenratanakul S. Acute respiratory social media for infl uenza surveillance. Adv Exp Med Biol illness in patients with COPD and the eff ectiveness of 2010; 680:559–564. infl uenza vaccination: a randomized controlled study. Chest 15. Mandell LA, Wunderink RG, Anzueto A, Bartlett JG, 2004; 125:2011–2020. Campbell GD, Dean NC, et al. Infectious Diseases Society 26. Monto AS. Seasonal infl uenza and vaccination coverage. of America/American Th oracic Society consensus guidelines Vaccine 2010; 28 Suppl 4:D33–44. on the management of community-acquired pneumonia in 27. Ginde AA, Tsai CL, Blanc PG, Camargo CA, Jr. Positive adults. Clin Infect Dis 2007; 44 Suppl 2:S27–72. predictive value of ICD-9-CM codes to detect acute 16. Wong ZX, Jones JE, Anderson GP, Gualano RC. Oseltamivir exacerbation of COPD in the emergency department. Jt Comm treatment of mice before or after mild infl uenza infection J Qual Patient Saf 2008; 34:678–680. reduced cellular and cytokine infl ammation in the lung. 28. Marsden-Haug N, Foster VB, Gould PL, Elbert E, Wang H, Infl uenza Other Respi Viruses 2011; 5(5):343–350. Pavlin JA . Code-based syndromic surveillance for infl uenzalike 17. Vincken W, van Noord JA, Greefhorst AP, Bantje TA, Kesten S, illness by International Classifi cation of Diseases, Ninth Korducki L, et al. Improved health outcomes in patients with Revision. Emerg Infect Dis 2007;13:207–216. COPD during 1 yr’s treatment with tiotropium. Eur Respir J 29. Stein BD, Bautista A, Schumock GT, Lee TA, Charbeneau JT, 2002; 19:209–216. Lauderdale DS, et al. Th e validity of ICD-9-CM diagnosis codes 18. Casaburi R, Mahler DA, Jones PW, Wanner A, San PG, for identifying patients hospitalized for COPD exacerbations. ZuWallack RL, et al. A long-term evaluation of once-daily Chest 2012; 141(1):87–93. A supplemental fi gure is available in the online version of this article. Copyright © 2013 Informa Healthcare USA, Inc

Journal

COPD Journal of Chronic Obstructive Pulmonary Disease – Taylor & Francis

Published: Oct 1, 2013

Keywords: Chronic airflow obstruction; human influenza; epidemiology; hospitalization; disease exacerbation; chronic obstructive pulmonary disease; COPD exacerbation.

There are no references for this article.