Journals of Gerontology: Medical Sciences cite as: J Gerontol A Biol Sci Med Sci, 2018, Vol. 73, No. 7, 988 doi:10.1093/gerona/gly055 Advance Access publication April 25, 2018 Letter to the Editor 1 2 3 Jodi B. Segal, MD, MPH, Ravi Varadhan PhD, Michelle C. Carlson, PhD, and Jeremy D. Walston, MD 1 2 Johns Hopkins University School of Medicine – Department of Medicine, Baltimore, Maryland. Johns Hopkins University School of Medicine – Oncology Center, Baltimore, Maryland. Johns Hopkins University Bloomberg School of Public Health –Department of Mental Health, Baltimore, Maryland. Address correspondence to: Jodi Segal, MD, MPH, Johns Hopkins University School of Medicine – Medicine, Baltimore, MD 21205. E-mail: email@example.com Received: January 24, 2018; Editorial Decision Date: February 23, 2018 To the Editor, in selecting the best model (out of the six candidates) via bootstrap- We were pleased to see the work by Kim and colleagues, Measuring ping, the adjustment does not mitigate the problem of over-fitting Fairly in Medicare Data: Development and Validation of a Claims- since the lasso penalty appears not to have been chosen to minimize Based Frailty Index (1). We have previously published a claims-based prediction error. Hence their approach may result in over-fitting and frailty assessment tool that was based on the physical frailty pheno- yield a larger than necessary model. type rather than a comorbidity based frailty index approach (2). We Finally, Kim et al.’s index comprises 93 variables, including those also externally validated our frailty tool in the National Health and that are claims for durable medical equipment that are less ubiqui- Aging Trends Study (3). The prediction characteristics (eg, c-statistic, tously available in data sets for research. On the other hand, our hazard ratios for various adverse events) from the validation exer- index uses less than 20% of the number of items (just 18) all of which cise in National Health and Aging Trends Study (NHATS) were are known from inpatient or outpatient claims, plus 3 demographic remarkably close to the prediction in the original training data in the variables. We believe that the parsimony of our index is primarily due Cardiovascular Health Study. We were pleased to see that Kim et al.’s to two factors: anchoring on a well characterized frailty phenotype index also predicted well the clinical outcomes of interest in the same and the use of cross-validation to control for over-fitting. For uptake data in the subsequent year. We look forward to seeing it validated in and use of the index, we maintain that parsimony is prized by users. different and diverse data sets. In summary, although there are important methodological dif- Although Kim et al.’s approach broadly resembles ours, there are ferences between our approaches, and a broad conceptual difference some important differences which raises some concern regarding the in the constructs of frailty approximated with claims data, we are methodology behind their claims-based tool. First, it is not clear if pleased to see that there is increased recognition of the value of a the claims-based frailty index is trying to approximate a survey-based claims-based frailty index for its many potential analytic uses. frailty index (SFI) or mortality risk or both. Given this, the overall approximation goal is unclear, leading to a lack of construct specificity. References Second, there are several methodological questions in regards to 1. Kim DH, Schneeweiss S, Glynn RJ, Lipsitz LA, Rockwood K, Avorn J. development of claims-based frailty index as a surrogate of survey- Measuring frailty in medicare data: development and validation of a claims- based frailty index. What is the need for applying lasso? How was based frailty index. J Gerontol A Biol Sci Med Sci. 2018;73:980–987. the lasso penalty chosen? Is survey-based frailty index a continuous doi:10.1093/gerona/glx229. variable between 0 and 1? If survey-based frailty index was used 2. Segal JB, Chang H-Y, Du Y, Walston JD, Carlson MC, Varadhan R. a continuous variable, there would be issues with parameter con- Development of a claims-based frailty indicator anchored to a well- straints in the regression. How was this addressed? established frailty phenotype. Med Care. 2017;55:716–722. doi:10.1097/ Third, in our work, we employed a 10-fold cross-validation strat- MLR.0000000000000729 egy to obtain the optimal regression model in terms of c-statistics 3. Segal JB, Huang J, Roth DL, Varadhan R. External validation of the rather than using the entire training sample to get the optimal model claims-based frailty index in the national health and aging trends study as in this study. Even though an adjustment for optimism was utilized cohort. Am J Epidemiol. 2017;186:745–747. doi:10.1093/aje/kwx257 © 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: firstname.lastname@example.org. Downloaded from https://academic.oup.com/biomedgerontology/article-abstract/73/7/988/4985509 by Ed 'DeepDyve' Gillespie user on 21 June 2018
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.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud