We appreciate the thoughtful comments and discussion by Dr. Varadhan and his colleagues on our recent work to develop and validate a claims-based frailty index (CFI) according to the deficit-accumulation approach using the Medicare Current Beneficiary Survey data (1). They also compared our CFI with their index that was developed using the frailty phenotype as an anchor from Medicare data (2). Details of our analyses had been part of the original submission, but they were simplified in our final paper due to space constraint and readability. Therefore, we welcome this opportunity to provide additional explanations about our analytic approach. The first point raised by Varadhan et al. was about construct specificity of our CFI. Our goal was to quantify frailty from claims data based on the deficit-accumulation approach. Mortality prediction was one of the several evaluations that we performed after the model was developed. As previously outlined (3), we considered two approaches to create a CFI. One approach was to estimate a CFI by counting the conditions defined using diagnosis codes, procedure codes, and health care service codes in claims data according to the standard deficit-accumulation approach. The key assumption was that receipt of health care services was a reliable proxy for health deficits. However, some may argue that receiving a treatment or health care service itself is not a health deficit; certain health deficits, such as memory loss, depression, or slow gait, cannot be accurately measured from routine health care data. Thus, we considered an alternative approach to create a standard deficit-accumulation frailty index from a survey—where deficits are directly measured from participants—and develop a function that approximates the survey-based frailty index (SFI) from claims data using a regression model. This is analogous to the method Varadhan et al. used, except that our anchor was the SFI instead of phenotypic frailty (2). Between our first approach (counting deficits directly from claims data) and second approach (approximating a SFI using a regression model), the latter achieved higher discrimination for mortality prediction (C statistics: 0.76–0.77) than the former approach (C statistics: 0.71–0.73) (Supplementary Table 2 in our paper (1)). Varadhan et al. asked several methodological questions about our CFI development. To develop a function to approximate the SFI (range: 0.000–0.705), we fitted a linear regression model. Prediction of a continuous frailty index can offer advantages over prediction of a dichotomous frailty phenotype because classifying individuals into frail versus nonfrail may result in loss of information on the severity of frailty. Quantifying the severity of frailty can be more useful for identifying populations at high risk for poor health outcomes and for adjustment for confounding by frailty in claims-based comparative effectiveness and safety studies. Although we aggregated over 30,000 individual codes in claims data into a smaller number of meaningful groups of diseases, procedures, and health care services and applied the prevalence threshold before fitting a linear model, the number of candidate variables was large (over 600 variables) and over-fitting was a concern. Therefore, the lasso algorithm was applied using the mean absolute error between the observed SFI and the model-predicted SFI as the loss function. The optimal lasso penalty was chosen based on this loss function using the “1-standard error rule” from a 10-fold cross-validation in the development sample (4). Since the model-predicted SFI ranged from 0.039 to 0.599, which was slightly narrower than the range of the observed SFI, we did not encounter any parameter constraint issues. After developing the model, we evaluated the predictive ability of our model for mortality and other adverse health outcomes (Table 3, Supplementary Tables 2 and 4 in our paper (1)) using the bootstrap C statistics in two separate years of data (2006–2007 and 2011–2012). Finally, Varadhan et al. commented that our CFI contains a larger number of variables from various data sources than their index that predicted the phenotypic frailty based on age, sex, race, the Charlson comorbidity index, and 17 conditions from inpatient and outpatient claims (2). While requiring additional datasets (eg, Current Procedural Terminology codes, Healthcare Common Procedure Coding System codes that include durable medical equipment codes) may limit transportability of our CFI to databases outside the United States, our results showed that 6 of the 10 strongest predictors of SFI were derived from durable medical equipment codes. Parsimony is always important but not at the cost of accuracy. Better discrimination is arguably more important than parsimony, especially when additional data are available for the entire Medicare population and can be obtained at no extra costs. Nonetheless, we agree that the usability of our CFI may be less than that of a shorter index. We are currently testing a SAS macro, which will be made available for users in near future. REFERENCES 1. Kim DH, Schneeweiss S, Glynn RJ, Lipsitz LA, Rockwood K, Avorn J. Measuring frailty in medicare data: development and validation of a claims-based frailty index. J Gerontol A Biol Sci Med Sci. 2017. doi:10.1093/gerona/glx229. [Epub ahead of print] . 2. Segal JB, Chang H-Y, Du Y, Walston JD, Carlson MC, Varadhan R. Development of a claims-based frailty indicator anchored to a well-established frailty phenotype. Med Care . 2017; 55: 716– 722. doi: 10.1097/MLR.0000000000000729 Google Scholar CrossRef Search ADS PubMed 3. Kim DH, Schneeweiss S. Measuring frailty using claims data for pharmacoepidemiologic studies of mortality in older adults: evidence and recommendations. Pharmacoepidemiol Drug Saf . 2014; 23: 891– 901. doi: 10.1002/pds.3674 Google Scholar CrossRef Search ADS PubMed 4. Tibshirani R. Regression shrinkage and selection via the lasso. J Roy Statist Soc Ser B . 1996; 58: 267– 288. © The Author(s) 2018. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: email@example.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)
The Journals of Gerontology Series A: Biomedical Sciences and Medical Sciences – Oxford University Press
Published: Apr 25, 2018
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
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.
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.
“Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”Daniel C.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud
“I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”@deepthiw
“My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”@JoseServera