pith. sign in

arxiv: 2605.20154 · v2 · pith:NURMAQKNnew · submitted 2026-05-19 · 📊 stat.ME · stat.AP

Component over Composite: Mitigating Type I Error Inflation when Imputing "Days Alive and at Home"

Mia S. Tackney , Sarah Dawson , Letao Yuan , Dominique-Laurent Couturier , Sofia S. Villar This is my paper

Pith reviewed 2026-05-21 07:18 UTC · model grok-4.3

classification 📊 stat.ME stat.AP
keywords missing datamultiple imputationcomposite outcomeDays Alive and at Hometype I errorclinical trialspredictive mean matchingMann-Whitney-Wilcoxon test
0
0 comments X

The pith

Imputing the separate components of Days Alive and at Home controls type I error better than imputing the composite directly.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines methods for handling missing data in the Days Alive and at Home outcome, which combines initial hospital stay, post-discharge care time, and mortality into a single measure for clinical trials. These components often have different missingness rates and mechanisms, so the authors test complete-case analysis, multiple imputation of the final composite score, and multiple imputation applied to each component before forming the composite. Simulations drawn from the NOTACS trial show that imputing at the component level keeps type I error rates appropriate and preserves power for the Mann-Whitney-Wilcoxon test, whereas imputing the composite with predictive mean matching tends to inflate the false-positive rate.

Core claim

Through a simulation study motivated by the NOTACS trial, we compare several methods of handling missing data, including complete case analysis, MI of the composite, and MI of the components when the primary analysis is a Mann-Whitney-Wilcoxon test. MI on the component level has good properties in terms of type I error control and power. We caution against the use of MI on the composite level with Predictive Mean Matching, which can lead to type I error inflation.

What carries the argument

Multiple imputation applied to the individual components of the DAH outcome before they are combined, which respects each component's distinct distribution and missing-data mechanism.

If this is right

  • Component-level multiple imputation keeps type I error rates at the nominal level for the Mann-Whitney-Wilcoxon test on DAH.
  • Direct imputation of the composite DAH value with predictive mean matching risks inflating type I error.
  • Complete-case analysis discards incomplete records and can reduce power compared with component imputation.
  • Future imputation methods should be developed to match more complex definitions of the DAH outcome.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The component-imputation approach may extend usefully to other composite endpoints that combine continuous lengths of stay with binary mortality indicators.
  • Trial protocols that adopt DAH should pre-specify component-level imputation and include sensitivity analyses for the missing-at-random assumption.
  • Repeating the simulation across additional trial datasets would clarify how sensitive the recommendation is to the specific missingness structure.

Load-bearing premise

The missing-data patterns and correlations among components observed in the NOTACS trial are representative of those in other trials that will use the DAH outcome.

What would settle it

A new simulation that draws missingness rates, mechanisms, and component correlations from a different real trial dataset using DAH, then checks whether component-level imputation still controls type I error while composite-level imputation with predictive mean matching inflates it.

Figures

Figures reproduced from arXiv: 2605.20154 by Dominique-Laurent Couturier, Letao Yuan, Mia S. Tackney, Sarah Dawson, Sofia S. Villar.

Figure 1
Figure 1. Figure 1: Colour-coded missingness status in location and support data from the 47 participants with missing data [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Top panel: Histogram of the DAH90 observations of a subset of 200 participants from the interim analysis [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Simulation results: Type I error (left) under the Null scenario and Power (right) under the Alternative scenario [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
read the original abstract

Background: Days Alive and at Home (DAH) over a pre-defined follow-up period is a novel post-intervention composite outcome that combines data from at least three components: (i) initial length of hospital stay, (ii) length of total readmissions or other post-discharge care and (iii) mortality. Missing values bring unique challenges to the analysis of trials with the DAH outcome as the three components may have different rates of missingness caused by distinct missing data mechanisms. Current approaches define DAH as missing if any of the components are missing, and proceed with complete cases or Multiple Imputation (MI) of the composite. Methods: Through a simulation study motivated by the NOTACS trial, we compare several methods of handling missing data, including complete case analysis, MI of the composite, and MI of the components when the primary analysis is a Mann-Whitney-Wilcoxon test. Results: MI on the component level has good properties in terms of type I error control and power. We caution against the use of MI on the composite level with Predictive Mean Matching, which can lead to type I error inflation. Conclusions: Given the complex distributional characteristics of DAH, naive approaches such as defining missingness on the composite level and directly imputing the composite with Predictive Mean Matching, can lead to type I error inflation. Imputing on the component level is recommended, suggested future work included imputation approaches that are compatible with more complex definitions of DAH, as well as recommendations for sensitivity analyses to the Missing at Random assumption.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript reports a simulation study motivated by the NOTACS trial comparing methods for handling missing data when analyzing the Days Alive and at Home (DAH) composite outcome. The authors evaluate complete-case analysis, multiple imputation applied directly to the composite DAH, and multiple imputation applied separately to the three components (initial length of stay, readmission/post-discharge days, and mortality) before constructing DAH, with the primary analysis being a Mann-Whitney-Wilcoxon test. The central finding is that component-level imputation maintains nominal type I error control and reasonable power, whereas composite-level imputation using predictive mean matching produces type I error inflation under the simulated conditions.

Significance. If the operating characteristics hold, this work supplies practical, actionable guidance for trialists using DAH endpoints, which are increasingly adopted in perioperative and critical-care studies. The simulation design is a clear strength: it is explicitly motivated by an existing trial, compares relevant methods, reports both type I error and power, and avoids post-hoc data exclusions. The explicit caution against composite-level PMM is a concrete contribution that can directly inform analysis plans. The authors also flag the need for future work on more complex DAH definitions and sensitivity analyses to the MAR assumption, which appropriately tempers the scope of the recommendations.

minor comments (2)
  1. [Results] Results section (tables reporting type I error): Monte Carlo standard errors should be reported alongside the empirical type I error rates so readers can judge the simulation precision directly.
  2. [Discussion] Discussion: The interaction between imputation method and the observed correlation structure among DAH components could be discussed in greater detail to help readers anticipate performance in trials with different dependence patterns.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive review and recommendation to accept the manuscript. We appreciate the recognition of the simulation study's motivation from the NOTACS trial, its clear design, and the practical implications for handling missing data in DAH endpoints.

Circularity Check

0 steps flagged

No circularity: empirical simulation study with independent Monte Carlo results

full rationale

This is an empirical simulation study whose type I error and power conclusions are generated directly from Monte Carlo replications under missingness mechanisms and correlations calibrated to the NOTACS trial. No algebraic identity, fitted parameter, or self-citation is reused as a prediction; the central comparison (component-level MI versus composite-level PMM) is evaluated by direct simulation output rather than by construction or by a load-bearing self-reference. The work is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper relies on the standard missing-at-random assumption required for multiple imputation validity and on the simulation design faithfully reproducing the NOTACS trial's missingness rates and correlations; no new entities are postulated and no parameters are fitted to produce the headline result.

axioms (1)
  • domain assumption Missing at random (MAR) mechanism governs the missingness in the three DAH components
    Standard assumption invoked for the validity of multiple imputation; stated in the methods description of the simulation.

pith-pipeline@v0.9.0 · 5827 in / 1339 out tokens · 29001 ms · 2026-05-21T07:18:13.535793+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages · 1 internal anchor

  1. [1]

    URLhttps://doi.org/10.1080/02664763.2024.2418473

    doi: 10.1080/02664763.2024.2418473. URLhttps://doi.org/10.1080/02664763.2024.2418473. Max Bell, Lars I. Eriksson, Tobias Svensson, Linn Hallqvist, Fredrik Granath, Jennifer Reilly, and Paul S. Myles. Days at home after surgery: An integrated and efficient outcome measure for clinical trials and quality assurance. EClinicalMedicine, 11:18–26,

    work page doi:10.1080/02664763.2024.2418473 2024

  2. [2]

    Mabel Chung, Zaid I

    doi: 10.1016/j.eclinm.2019.04.011. Mabel Chung, Zaid I. Almarzooq, Archana Tale, Yang Song, Issa J. Dahabreh, Dhruv S. Kazi, Suzanne J. Baron, and Robert W. Yeh. Days at home after transcatheter mitral valve repair versus medical therapy alone in heart failure. Journal of the American Heart Association, 14, 1

    work page doi:10.1016/j.eclinm.2019.04.011 2019

  3. [3]

    doi: 10.1161/JAHA.124.038401

    ISSN 20479980. doi: 10.1161/JAHA.124.038401. S. N. Dawson, Y . D. Chiu, A. A. Klein, et al. Effect of high-flow nasal therapy on patient-centred outcomes in patients at high risk of postoperative pulmonary complications after cardiac surgery: a statistical analysis plan for notacs, a multicentre adaptive randomised controlled trial.Trials, 23(1):699,

    work page doi:10.1161/jaha.124.038401

  4. [4]

    URL https://doi.org/10.1186/s13063-022-06607-z

    doi: 10.1186/s13063-022-06607-z. URL https://doi.org/10.1186/s13063-022-06607-z. Sarah N Dawson, Yi-Da Chiu, Andrew A Klein, Melissa Earwaker, Sofia S Villar, and NOTACS study team. Effect of high-flow nasal therapy on patient-centred outcomes in patients at high risk of postoperative pulmonary complications after cardiac surgery: update to the statistica...

    work page doi:10.1186/s13063-022-06607-z

  5. [5]

    e-Collection

    doi: 10.1080/02664763.2021.1895087. e-Collection

    work page doi:10.1080/02664763.2021.1895087 2021

  6. [6]

    URL https: //bmjopen.bmj.com/content/7/8/e015828

    doi: 10.1136/bmjopen-2017-015828. URL https: //bmjopen.bmj.com/content/7/8/e015828. Paul S. Myles, Jan M. Dieleman, Andrew Forbes, Stephane Heritier, and Julian A. Smith. Dexamethasone for cardiac surgery trial (decs-ii): Rationale and a novel, practice preference-randomized consent design.American Heart Journal, 204:52–57,

    work page doi:10.1136/bmjopen-2017-015828 2017

  7. [7]

    doi: https://doi.org/10.1016/j.ahj.2018.06.008

    ISSN 0002-8703. doi: https://doi.org/10.1016/j.ahj.2018.06.008. URL https: //www.sciencedirect.com/science/article/pii/S000287031830190X. Aidan G. O’Keeffe, Daniel M. Farewell, Brian D. M. Tom, and Vernon T. Farewell. Multiple imputation of missing com- posite outcomes in longitudinal data.Statistical Biosciences, 8(2):310–332,

    work page doi:10.1016/j.ahj.2018.06.008 2018

  8. [8]

    URLhttps://doi.org/10.1007/s12561-016-9146-z

    doi: 10.1007/s12561-016-9146-z. URLhttps://doi.org/10.1007/s12561-016-9146-z. Tra My Pham, Ian R. White, Brennan C. Kahan, Tim P. Morris, Simon J. Stanworth, and Gordon Forbes. A comparison of methods for analyzing a binary composite endpoint with partially observed components in randomized controlled trials.Statistics in Medicine, 40:6634–6650, 12

    work page doi:10.1007/s12561-016-9146-z

  9. [9]

    doi: 10.1002/sim.9203

    ISSN 10970258. doi: 10.1002/sim.9203. Donald B. Rubin.Multiple Imputation for Nonresponse in Surveys. John Wiley & Sons, New York,

    work page doi:10.1002/sim.9203

  10. [10]

    doi: 10.1186/1471-2288-12-46

    ISSN 1471-2288. doi: 10.1186/1471-2288-12-46. URL https://doi.org/10.1186/1471-2288-12-46 . Ralph Shackleton, Sarah V ollam, Stephen Gerry, Akshay Shah, and David M. Griffith. Reporting of hospital-free days as an outcome measure in critical care trials: A systematic review.Critical Care Medicine, 53(12):e2686–e2697, December

    work page doi:10.1186/1471-2288-12-46

  11. [11]

    doi: 10.1097/CCM.0000000000006858. S. Shetty, M. Duckworth, R. Norman, J. Affandi, S. Dawson, J. Fox-Rushby, NOTACS investigators, NOTACS Trial Steering Committee, and NOTACS DMEC Committee. The international trial of nasal oxygen therapy after cardiac surgery (notacs) in patients at high risk of postoperative pulmonary complications: Economic evaluation ...

    work page doi:10.1097/ccm.0000000000006858 2026

  12. [12]

    URL https://doi.org/10.1371/journal.pone.0311861

    doi: 10.1371/journal.pone.0311861. URL https://doi.org/10.1371/journal.pone.0311861. Thomas R Sullivan, Ian R White, Amy B Salter, Philip Ryan, and Katherine J Lee. Should multiple imputation be the method of choice for handling missing data in randomized trials?Statistical Methods in Medical Research, 27: 2610–2626, 9

    work page doi:10.1371/journal.pone.0311861

  13. [13]

    doi: 10.1177/0962280216683570

    ISSN 0962-2802. doi: 10.1177/0962280216683570. Paul T. von Hippel. How to impute interactions, squares, and other transformed variables.Sociological Methodology, 39:265–291,

    work page doi:10.1177/0962280216683570

  14. [14]

    doi: 10.1111/j.1467-9531.2009.01245.x. Ian R. White, Patrick Royston, and Angela M. Wood. Multiple imputation using chained equations: Issues and guidance for practice.Statistics in Medicine, 30:377–399, 2

    work page doi:10.1111/j.1467-9531.2009.01245.x 2009

  15. [15]

    Days Alive and at Home

    ISSN 0277-6715. doi: 10.1002/sim.4067. Letao Yuan, Sofía S. Villar, and Dominique-Laurent Couturier. Beyond the composite: Enhancing trial analysis through a divide & conquer approach to “Days Alive and at Home": Insights from the NOTACS trial,

    work page doi:10.1002/sim.4067

  16. [16]

    URL https://arxiv.org/abs/2605.16606. 12

    work page internal anchor Pith review Pith/arXiv arXiv