LASSO-Based Survival Prediction Modeling with Multiply Imputed Data: A Case Study in Tuberculosis Mortality Prediction.

Using health administrative datasets for developing prediction models is always challenging due to missing values in key predictors. Multiple imputation has been recommended to deal with missing predictor values. However, predicting survival outcomes using regularized regression, for example, Cox-LA...

Descripción completa

Detalles Bibliográficos
Publicado en:American Statistician Vol. 80; no. 1; pp. 77 - 89
Autores principales: Hossain, Md. Belal, Sadatsafavi, Mohsen, Johnston, James C., Wong, Hubert, Cook, Victoria J., Karim, Mohammad Ehsanul
Formato: Artículo
Publicado: Taylor & Francis Ltd Feb2026
Materias:
Acceso en línea:Ver este registro en EBSCOhost
fields @attributes:
  recordID: 1
pdfLink:
plink: https://search.ebscohost.com/login.aspx?direct=true&db=ssf&AN=191630181&site=ehost-live
header:
  @attributes:
    shortDbName: ssf
    uiTerm: 191630181
    longDbName: Social Sciences Full Text (H.W. Wilson)
    uiTag: AN
  controlInfo:
    bkinfo:
    jinfo:
      jid:
        00031305
        STT
      jtl: American Statistician
      issn: 00031305
      maglogo: Y
    pubinfo:
      dt: Feb2026
      vid: 80
      iid: 1
      pid: 377
      pub: Taylor & Francis Ltd
    artinfo:
      ui:
        191630181
        10.1080/00031305.2025.2526545
      ppf: 77
      ppct: 12
      formats:
      tig:
        atl: LASSO-Based Survival Prediction Modeling with Multiply Imputed Data: A Case Study in Tuberculosis Mortality Prediction.
      aug:
        au:
          Hossain, Md. Belal
          Sadatsafavi, Mohsen
          Johnston, James C.
          Wong, Hubert
          Cook, Victoria J.
          Karim, Mohammad Ehsanul
        affil:
          School of Population and Public Health, University of British Columbia, Vancouver, BC, Canada
          Centre for Advancing Health Outcomes, St. Paul's Hospital, Vancouver, BC, Canada
          Faculty of Pharmaceutical Sciences, University of British Columbia, Vancouver, BC, Canada
          British Columbia Centre for Disease Control, Vancouver, BC, Canada
          Department of Medicine, University of British Columbia, Vancouver, BC, Canada
      su:
        Multiple imputation (Statistics)
        Tuberculosis mortality
        Survival analysis (Biometry)
        Regression analysis
        Statistical models
        Proportional hazards models
        Regularization parameter
      sug:
        subj:
          Multiple imputation (Statistics)
          Tuberculosis mortality
          Survival analysis (Biometry)
          Regression analysis
          Statistical models
          Proportional hazards models
          Regularization parameter
      keyword:
        Internal validation
        LASSO
        Multiple imputation
        Prediction
        Survival outcome
        Internal validation
        LASSO
        Multiple imputation
        Prediction
        Survival outcome
      ab: Using health administrative datasets for developing prediction models is always challenging due to missing values in key predictors. Multiple imputation has been recommended to deal with missing predictor values. However, predicting survival outcomes using regularized regression, for example, Cox-LASSO, faces limitations as these methods are incompatible with pooling model outputs from multiple imputed data using Rubin's rule. In this study, we explored the performance of three statistical methods in developing prediction models with Cox-LASSO on multiply imputed data: prediction average, performance average, and stacked. We considered two hyperparameter selection techniques: minimum-lambda that gives the minimum cross-validated prediction error and 1SE-lambda that selects more parsimonious models. We also conducted plasmode simulations with varying the events per parameter. The stacked approach provided the most robust predictions in our case study of predicting tuberculosis mortality and simulations, producing a time-dependent c-statistic of 0.93 and a well-calibrated calibration plot. The 1SE-lambda technique resulted in underfitting of the models in most scenarios, both in case study and simulation. Our findings advocate the stacked method with minimum-lambda as an effective technique for combining LASSO-based prediction outputs from multiply imputed data. We shared reproducible R codes for future researchers to facilitate the adoption of these methodologies in their research.
      pubtype: Academic Journal
      doctype: Article
      src: R
    language: English
    refInfo:
    copyright:
      @attributes:
        flag: N
    holdings:
      @attributes:
        islocal: N