Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques

Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques. Franklin, J., M., Shrank, W., H., Lii, J., Krumme, A., K., Matlin, O., S., Brennan, T., A., & Choudhry, N., K. Health Services Research, 51(1):220-239, Wiley/Blackwell (10.1111), 2, 2016.

Website abstract bibtex

Objective. Despite the proliferation of databases with increasingly rich patient data,prediction of medication adherence remains poor. We proposed and evaluatedapproaches for improved adherence prediction.Data Sources. We identiﬁed Medicare beneﬁciaries who received prescription drugcoverage through CVS Caremark and initiated a statin.Study Design. A total of 643 variables were identiﬁed at baseline from prior claimsand linked Census data. In addition, we identiﬁed three postbaseline predictors, indica-tors of adherence to statins during each of the ﬁrst 3 months of follow-up. We estimated10 models predicting subsequent adherence, using logistic regression and boostedlogistic regression, a nonparametric data-mining technique. Models were also esti-mated within strata deﬁned by the index days supply.Results. In 77,703 statin initiators, prediction using baseline variables only was poorwith maximum cross-validated C-statistics of 0.606 and 0.577 among patients withindex supply ≤30 days and >30 days, respectively. Using only indicators of initial sta-tin adherence improved prediction accuracy substantially among patients with shorterinitial dispensings (C = 0.827/0.518), and, when combined with investigator-speciﬁ edvariables, prediction accuracy was further improved (C = 0.842/0.596).Conclusions. Observed adherence immediately after initiation predicted futureadherence for patients whose initial dispensings were relatively short.

@article{
 title = {Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques},
 type = {article},
 year = {2016},
 identifiers = {[object Object]},
 keywords = {Adherence,boosting,comparative effectiveness,epidemiologic methods,prediction},
 pages = {220-239},
 volume = {51},
 websites = {http://doi.wiley.com/10.1111/1475-6773.12310},
 month = {2},
 publisher = {Wiley/Blackwell (10.1111)},
 id = {3c672fa0-f69b-339d-823e-691024920341},
 created = {2018-06-26T08:46:24.224Z},
 accessed = {2018-06-26},
 file_attached = {false},
 profile_id = {35f49df4-e9cf-391e-a76a-8e9e316c42aa},
 group_id = {ee11d323-dd56-3515-8742-9b1e2f0496f7},
 last_modified = {2018-06-30T13:47:49.679Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 folder_uuids = {1a7012b8-c568-4f80-b7bd-80d07dcffde5,0f486643-3d08-4632-8e8e-e53433d903a9},
 private_publication = {false},
 abstract = {Objective. Despite the proliferation of databases with increasingly rich patient data,prediction of medication adherence remains poor. We proposed and evaluatedapproaches for improved adherence prediction.Data Sources. We identiﬁed Medicare beneﬁciaries who received prescription drugcoverage through CVS Caremark and initiated a statin.Study Design. A total of 643 variables were identiﬁed at baseline from prior claimsand linked Census data. In addition, we identiﬁed three postbaseline predictors, indica-tors of adherence to statins during each of the ﬁrst 3 months of follow-up. We estimated10 models predicting subsequent adherence, using logistic regression and boostedlogistic regression, a nonparametric data-mining technique. Models were also esti-mated within strata deﬁned by the index days supply.Results. In 77,703 statin initiators, prediction using baseline variables only was poorwith maximum cross-validated C-statistics of 0.606 and 0.577 among patients withindex supply ≤30 days and >30 days, respectively. Using only indicators of initial sta-tin adherence improved prediction accuracy substantially among patients with shorterinitial dispensings (C = 0.827/0.518), and, when combined with investigator-speciﬁ edvariables, prediction accuracy was further improved (C = 0.842/0.596).Conclusions. Observed adherence immediately after initiation predicted futureadherence for patients whose initial dispensings were relatively short.},
 bibtype = {article},
 author = {Franklin, Jessica M. and Shrank, William H. and Lii, Joyce and Krumme, Alexis K. and Matlin, Olga S. and Brennan, Troyen A. and Choudhry, Niteesh K.},
 journal = {Health Services Research},
 number = {1}
}

Downloads: 0

{"_id":"jbLyTuJizkWfiD2Tf","bibbaseid":"franklin-shrank-lii-krumme-matlin-brennan-choudhry-observingversuspredictinginitialpatternsoffillingpredictlongtermadherencemoreaccuratelythanhighdimensionalmodelingtechniques-2016","downloads":0,"creationDate":"2018-06-30T13:35:33.799Z","title":"Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques","author_short":["Franklin, J., M.","Shrank, W., H.","Lii, J.","Krumme, A., K.","Matlin, O., S.","Brennan, T., A.","Choudhry, N., K."],"year":2016,"bibtype":"article","biburl":null,"bibdata":{"title":"Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques","type":"article","year":"2016","identifiers":"[object Object]","keywords":"Adherence,boosting,comparative effectiveness,epidemiologic methods,prediction","pages":"220-239","volume":"51","websites":"http://doi.wiley.com/10.1111/1475-6773.12310","month":"2","publisher":"Wiley/Blackwell (10.1111)","id":"3c672fa0-f69b-339d-823e-691024920341","created":"2018-06-26T08:46:24.224Z","accessed":"2018-06-26","file_attached":false,"profile_id":"35f49df4-e9cf-391e-a76a-8e9e316c42aa","group_id":"ee11d323-dd56-3515-8742-9b1e2f0496f7","last_modified":"2018-06-30T13:47:49.679Z","read":false,"starred":false,"authored":false,"confirmed":false,"hidden":false,"folder_uuids":"1a7012b8-c568-4f80-b7bd-80d07dcffde5,0f486643-3d08-4632-8e8e-e53433d903a9","private_publication":false,"abstract":"Objective. Despite the proliferation of databases with increasingly rich patient data,prediction of medication adherence remains poor. We proposed and evaluatedapproaches for improved adherence prediction.Data Sources. We identiﬁed Medicare beneﬁciaries who received prescription drugcoverage through CVS Caremark and initiated a statin.Study Design. A total of 643 variables were identiﬁed at baseline from prior claimsand linked Census data. In addition, we identiﬁed three postbaseline predictors, indica-tors of adherence to statins during each of the ﬁrst 3 months of follow-up. We estimated10 models predicting subsequent adherence, using logistic regression and boostedlogistic regression, a nonparametric data-mining technique. Models were also esti-mated within strata deﬁned by the index days supply.Results. In 77,703 statin initiators, prediction using baseline variables only was poorwith maximum cross-validated C-statistics of 0.606 and 0.577 among patients withindex supply ≤30 days and >30 days, respectively. Using only indicators of initial sta-tin adherence improved prediction accuracy substantially among patients with shorterinitial dispensings (C = 0.827/0.518), and, when combined with investigator-speciﬁ edvariables, prediction accuracy was further improved (C = 0.842/0.596).Conclusions. Observed adherence immediately after initiation predicted futureadherence for patients whose initial dispensings were relatively short.","bibtype":"article","author":"Franklin, Jessica M. and Shrank, William H. and Lii, Joyce and Krumme, Alexis K. and Matlin, Olga S. and Brennan, Troyen A. and Choudhry, Niteesh K.","journal":"Health Services Research","number":"1","bibtex":"@article{\n title = {Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques},\n type = {article},\n year = {2016},\n identifiers = {[object Object]},\n keywords = {Adherence,boosting,comparative effectiveness,epidemiologic methods,prediction},\n pages = {220-239},\n volume = {51},\n websites = {http://doi.wiley.com/10.1111/1475-6773.12310},\n month = {2},\n publisher = {Wiley/Blackwell (10.1111)},\n id = {3c672fa0-f69b-339d-823e-691024920341},\n created = {2018-06-26T08:46:24.224Z},\n accessed = {2018-06-26},\n file_attached = {false},\n profile_id = {35f49df4-e9cf-391e-a76a-8e9e316c42aa},\n group_id = {ee11d323-dd56-3515-8742-9b1e2f0496f7},\n last_modified = {2018-06-30T13:47:49.679Z},\n read = {false},\n starred = {false},\n authored = {false},\n confirmed = {false},\n hidden = {false},\n folder_uuids = {1a7012b8-c568-4f80-b7bd-80d07dcffde5,0f486643-3d08-4632-8e8e-e53433d903a9},\n private_publication = {false},\n abstract = {Objective. Despite the proliferation of databases with increasingly rich patient data,prediction of medication adherence remains poor. We proposed and evaluatedapproaches for improved adherence prediction.Data Sources. We identiﬁed Medicare beneﬁciaries who received prescription drugcoverage through CVS Caremark and initiated a statin.Study Design. A total of 643 variables were identiﬁed at baseline from prior claimsand linked Census data. In addition, we identiﬁed three postbaseline predictors, indica-tors of adherence to statins during each of the ﬁrst 3 months of follow-up. We estimated10 models predicting subsequent adherence, using logistic regression and boostedlogistic regression, a nonparametric data-mining technique. Models were also esti-mated within strata deﬁned by the index days supply.Results. In 77,703 statin initiators, prediction using baseline variables only was poorwith maximum cross-validated C-statistics of 0.606 and 0.577 among patients withindex supply ≤30 days and >30 days, respectively. Using only indicators of initial sta-tin adherence improved prediction accuracy substantially among patients with shorterinitial dispensings (C = 0.827/0.518), and, when combined with investigator-speciﬁ edvariables, prediction accuracy was further improved (C = 0.842/0.596).Conclusions. Observed adherence immediately after initiation predicted futureadherence for patients whose initial dispensings were relatively short.},\n bibtype = {article},\n author = {Franklin, Jessica M. and Shrank, William H. and Lii, Joyce and Krumme, Alexis K. and Matlin, Olga S. and Brennan, Troyen A. and Choudhry, Niteesh K.},\n journal = {Health Services Research},\n number = {1}\n}","author_short":["Franklin, J., M.","Shrank, W., H.","Lii, J.","Krumme, A., K.","Matlin, O., S.","Brennan, T., A.","Choudhry, N., K."],"urls":{"Website":"http://doi.wiley.com/10.1111/1475-6773.12310"},"bibbaseid":"franklin-shrank-lii-krumme-matlin-brennan-choudhry-observingversuspredictinginitialpatternsoffillingpredictlongtermadherencemoreaccuratelythanhighdimensionalmodelingtechniques-2016","role":"author","keyword":["Adherence","boosting","comparative effectiveness","epidemiologic methods","prediction"],"downloads":0},"search_terms":["observing","versus","predicting","initial","patterns","filling","predict","long","term","adherence","more","accurately","high","dimensional","modeling","techniques","franklin","shrank","lii","krumme","matlin","brennan","choudhry"],"keywords":["adherence","boosting","comparative effectiveness","epidemiologic methods","prediction"],"authorIDs":[]}