pith. sign in

arxiv: 2604.05283 · v1 · submitted 2026-04-07 · 📊 stat.ME

Truncation by death in the sufficient cause framework

Bronner P. Gon\c{c}alves , Eiji Yamamoto , Etsuji Suzuki This is my paper

Pith reviewed 2026-05-10 19:56 UTC · model grok-4.3

classification 📊 stat.ME
keywords sufficient cause frameworktruncation by deathsurvivor average causal effectprincipal stratificationcausal inferencerisk status typesbackground factorsselection bias
0
0 comments X

The pith

The crude comparison of outcomes among survivors mixes distinct susceptibility groups rather than estimating a causal effect.

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

This paper uses the sufficient cause framework to reinterpret truncation by death, the setting where the outcome cannot be observed for people who die before the study ends. It demonstrates that conditioning on observed survival produces a non-causal contrast because the surviving groups under different treatments contain different mixes of background-factor combinations that determine who is susceptible to death. The authors supply explicit expressions that write both the crude conditional estimand and the survivor average causal effect as functions of the population joint frequencies of those background factors. They also give conditions on the background factors under which the survivor average causal effect must be zero. Readers care because the reframing turns a common selection problem into a question about measurable risk profiles rather than an abstract bias.

Core claim

Under the sufficient cause framework, truncation by death occurs when individuals are susceptible to sufficient causes of death that render the outcome undefined. The crude estimand obtained by comparing outcomes conditional on observed survival is then a contrast between distinct risk status types, each defined by a unique pattern of susceptibility to the relevant sufficient causes. Both this crude estimand and the survivor average causal effect are expressed directly in terms of the population-level joint frequencies of the background factors of those sufficient causes. Conditions on the same background factors are stated under which the survivor average causal effect equals zero.

What carries the argument

Risk status types defined by patterns of susceptibility to sufficient causes, expressed through the joint frequencies of their background factors.

If this is right

  • The crude conditional comparison does not recover the survivor average causal effect.
  • Both the crude estimand and the survivor average causal effect can be written exactly using only the joint frequencies of background factors.
  • The survivor average causal effect is identically zero whenever the background factors satisfy the stated balance or independence conditions.
  • No additional assumptions about unmeasured common causes of survival and outcome are required beyond the structure already present in the sufficient cause model.

Where Pith is reading between the lines

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

  • Measuring or proxying the background factors in a study would allow direct calculation of the survivor average causal effect from observed frequencies.
  • The same frequency-based approach could be applied to other selection problems such as competing risks or loss to follow-up.
  • If the conditions for a zero survivor average causal effect hold, any remaining difference among survivors would point to model misspecification rather than a true treatment effect.

Load-bearing premise

The sufficient cause framework extends directly to truncation by death, with background factors alone fully defining risk status types and their joint frequencies without extra assumptions about unmeasured factors linking survival and the outcome.

What would settle it

A simulation or dataset in which the numerical value of the observed crude estimand fails to match the value predicted from the joint frequencies of the background factors under the sufficient cause model.

read the original abstract

The sufficient cause framework has been used for decades to improve our understanding of both basic and more complex causal concepts in epidemiology, such as mediation and interaction. Here, we make use of this framework to provide a description of truncation by death, in which the outcome of interest is undefined for individuals who die before the time of assessment at the end of follow-up. We explain the non-causal nature of the crude estimand that compares outcomes by treatment levels conditional on observed survival by showing that it corresponds to a comparison of distinct risk status types, which are defined based on the susceptibility to sufficient causes. Further, expressions for the crude estimand and for the survivor average causal effect, a causal estimand defined under the principal stratification approach, are provided in terms of population-level joint frequencies of the background factors of sufficient causes. Finally, we also describe conditions, based on background factors of sufficient causes, under which the survivor average causal effect is null. Our description of this problem, which studies truncation by death from a new perspective, might encourage further analyses of principal stratification-based estimands using sufficient causes.

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

2 major / 1 minor

Summary. The manuscript extends the sufficient cause framework to truncation by death. It argues that the crude estimand comparing outcomes by treatment conditional on observed survival is non-causal because it corresponds to a comparison of distinct risk status types defined by susceptibility to sufficient causes. Expressions for both the crude estimand and the survivor average causal effect (SACE) are provided in terms of population-level joint frequencies of the background factors of sufficient causes, along with conditions based on those background factors under which the SACE is null.

Significance. If the derivations hold without hidden assumptions, the work supplies a frequency-based re-expression of principal stratification concepts within the sufficient cause framework, which has previously clarified mediation and interaction. The explicit mapping to risk status types and the null conditions for SACE constitute a modest but concrete contribution that could prompt further cross-framework analyses in survival settings. The grounding in observable joint frequencies is a positive feature, as it avoids purely abstract potential-outcome notation.

major comments (2)
  1. [Modeling section on extension to truncation by death] The central extension of the sufficient cause framework to truncation by death (described in the main modeling section) rests on the claim that background factors alone suffice to define risk status types and their joint frequencies for both survival and the outcome. This mapping is load-bearing for the non-causal interpretation of the crude estimand and for the SACE expressions; the manuscript should explicitly state whether the joint distribution over the relevant U's is left unrestricted or implicitly requires independence or other restrictions, because truncation by death involves the joint law of potential survival indicators.
  2. [Section on expressions for crude estimand and SACE] In the section deriving expressions for the crude estimand and SACE, the reductions to comparisons of background-factor frequencies appear to follow by definition within the framework. To establish that the crude estimand is non-causal in a non-tautological sense, the paper should provide either a concrete numerical example showing bias that is invisible in the standard potential-outcomes notation or an explicit contrast with the usual principal-stratum derivation.
minor comments (1)
  1. [Abstract and introduction] The abstract and introduction would benefit from a brief statement of how the frequency-based null conditions differ from or complement existing monotonicity or no-interaction assumptions in the principal stratification literature.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive comments on our manuscript. We address each major comment below, indicating the revisions we plan to make to improve clarity and strengthen the presentation.

read point-by-point responses

  1. Referee: [Modeling section on extension to truncation by death] The central extension of the sufficient cause framework to truncation by death (described in the main modeling section) rests on the claim that background factors alone suffice to define risk status types and their joint frequencies for both survival and the outcome. This mapping is load-bearing for the non-causal interpretation of the crude estimand and for the SACE expressions; the manuscript should explicitly state whether the joint distribution over the relevant U's is left unrestricted or implicitly requires independence or other restrictions, because truncation by death involves the joint law of potential survival indicators.

    Authors: We agree that an explicit statement on this point is warranted to ensure the framework's generality is clear. In the revised manuscript, we will add a dedicated paragraph in the modeling section stating that the joint distribution over the background factors U is left completely unrestricted, permitting arbitrary dependence structures among the potential survival indicators. This generality is fundamental to the truncation-by-death setting and aligns with the use of population-level joint frequencies throughout the derivations. revision: yes

  2. Referee: [Section on expressions for crude estimand and SACE] In the section deriving expressions for the crude estimand and SACE, the reductions to comparisons of background-factor frequencies appear to follow by definition within the framework. To establish that the crude estimand is non-causal in a non-tautological sense, the paper should provide either a concrete numerical example showing bias that is invisible in the standard potential-outcomes notation or an explicit contrast with the usual principal-stratum derivation.

    Authors: We acknowledge the value of a concrete illustration to demonstrate the non-causal nature beyond the definitional mapping. We will incorporate a numerical example in the revised version of the section on expressions for the crude estimand and SACE. The example will use a small finite population with specified background-factor frequencies to show how the crude estimand compares distinct risk-status types (leading to bias relative to the SACE), and we will provide an explicit side-by-side contrast with the principal-stratification derivation to highlight the additional insight from the sufficient-cause perspective. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation extends external sufficient cause framework without reducing to self-inputs

full rationale

The paper maps the standard crude estimand (outcome comparison conditional on observed survival) and the survivor average causal effect onto risk status types and joint frequencies of background factors within the pre-existing sufficient cause framework. This is an explanatory correspondence rather than a self-definitional loop: the framework is cited as established (Rothman and others), the extension to truncation by death is stated as an assumption, and expressions are derived from that mapping without any fitted parameters, self-citations that bear the central load, or renaming of known results as new derivations. No equation or claim reduces by construction to its own inputs; the non-causal interpretation and null conditions follow from the framework's logic applied to principal strata, which remains externally grounded and falsifiable against standard truncation-by-death literature.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The analysis rests entirely on the established sufficient cause framework and principal stratification without new free parameters or invented entities; background factors are treated as given primitives whose joint frequencies define the estimands.

axioms (1)
  • domain assumption Outcomes result from combinations of background factors and sufficient causes, and this structure applies to modeling truncation by death and survival.
    Invoked throughout to define risk status types and express estimands in terms of joint frequencies.

pith-pipeline@v0.9.0 · 5495 in / 1338 out tokens · 114558 ms · 2026-05-10T19:56:07.914582+00:00 · methodology

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

38 extracted references · 38 canonical work pages

  1. [1]

    censoring

    Rubin, D.B., Causal inference through potential outcomes and principal stratification: application to studies with “censoring” due to death. Statistical Science, 2006. 21(3): p. 299-309

    work page 2006

  2. [2]

    Zhang, J.L. and D.B. Rubin, Estimation of Causal Effects via Principal Stratification When Some Outcomes are Truncated by “Death” . Journal of Educational and Behavioral Statistics, 2003. 28(4): p. 353-368

    work page 2003

  3. [3]

    Clin Trials, 2025: p

    Pham, T.M., et al., Estimating treatment effects in trials with outcome data truncated by death: A case study on aligning estimators with estimands. Clin Trials, 2025: p. 17407745251360645

    work page 2025

  4. [4]

    Use, Editor

    ICH harmonised guideline addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials ICH E9(R1), I.C.f.H.o.T.R.f.P .f.H. Use, Editor. 2019

    work page 2019

  5. [5]

    Journal of the American Statistical Association, 2005

    Rubin, D.B., Causal Inference Using Potential Outcomes: Design, Modeling, Decisions. Journal of the American Statistical Association, 2005. 100(469): p. 322- 331

    work page 2005

  6. [6]

    Epidemiology, 2015

    Suzuki, E., Generalized causal measure: the beauty lies in its generality. Epidemiology, 2015. 26(4): p. 490-5

    work page 2015

  7. [7]

    Frangakis, C.E. and D.B. Rubin, Principal stratification in causal inference. Biometrics, 2002. 58(1): p. 21-9

    work page 2002

  8. [8]

    Journal of the American Statistical Association, 2023

    Stensrud, M.J., et al., Conditional Separable Effects. Journal of the American Statistical Association, 2023. 118(544): p. 2671-2683

    work page 2023

  9. [9]

    Little, R.J. and D.B. Rubin, Causal effects in clinical and epidemiological studies via potential outcomes: concepts and analytical approaches. Annu Rev Public Health, 2000. 21: p. 121-45

    work page 2000

  10. [10]

    Truncation by Death

    Goncalves, B.P . and E. Suzuki, Effect Modification in Settings with "Truncation by Death" . Epidemiology, 2025. 36(3): p. 374-380

    work page 2025

  11. [11]

    Taguri, and Y

    Chiba, Y ., M. Taguri, and Y . Uemura, On the Identification of the Survivor Average Causal Effect. Journal of Biometrics & Biostatistics, 2011. 2

    work page 2011

  12. [12]

    Chiba, Y . and T.J. VanderWeele, A simple method for principal strata effects when the outcome has been truncated due to death. Am J Epidemiol, 2011. 173(7): p. 745-51

    work page 2011

  13. [13]

    Stat Med, 2014

    Tchetgen Tchetgen, E.J., Identification and estimation of survivor average causal effects. Stat Med, 2014. 33(21): p. 3601-28

    work page 2014

  14. [14]

    Zhou, and T.S

    Wang, L., X.H. Zhou, and T.S. Richardson, Identification and estimation of causal effects with outcomes truncated by death. Biometrika, 2017. 104(3): p. 597-612

    work page 2017

  15. [15]

    Suzuki, E., Unraveling Causality: Innovations in Epidemiologic Methods. JMA J,

    work page

  16. [16]

    Greenland, S. and B. Brumback, An overview of relations among causal modelling methods. Int J Epidemiol, 2002. 31(5): p. 1030-7

    work page 2002

  17. [17]

    Eur J Epidemiol, 2006

    Flanders, W.D., On the relationship of sufficient component cause models with potential outcome (counterfactual) models. Eur J Epidemiol, 2006. 21(12): p. 847- 53

    work page 2006

  18. [18]

    Lash, T.L., et al., Modern Epidemiology . 4th ed. 2020: Lippincott Williams & Wilkins

    work page 2020

  19. [19]

    Am J Epidemiol, 1976

    Rothman, K.J., Causes. Am J Epidemiol, 1976. 104(6): p. 587-92

    work page 1976

  20. [20]

    Yamamoto, and T

    Suzuki, E., E. Yamamoto, and T. Tsuda, On the link between sufficient -cause model and potential-outcome model. Epidemiology, 2011. 22(1): p. 131-2

    work page 2011

  21. [21]

    Yamamoto, and T

    Suzuki, E., E. Yamamoto, and T. Tsuda, On the relations between excess fraction, attributable fraction, and etiologic fraction. Am J Epidemiol, 2012. 175(6): p. 567- 75

    work page 2012

  22. [22]

    Tsuda, and E

    Suzuki, E., T. Tsuda, and E. Yamamoto, Covariate balance for no confounding in the sufficient-cause model. Ann Epidemiol, 2018. 28(1): p. 48-53.e2

    work page 2018

  23. [23]

    Suzuki, E. and E. Yamamoto, Marginal Sufficient Component Cause Model: An Emerging Causal Model With Merits? Epidemiology, 2021. 32(6): p. 838-845

    work page 2021

  24. [24]

    Suzuki, E. and E. Yamamoto, Strength in causality: discerning causal mechanisms in the sufficient cause model. Eur J Epidemiol, 2021. 36(9): p. 899-908

    work page 2021

  25. [25]

    Eur J Epidemiol, 2009

    VanderWeele, T.J., Mediation and mechanism. Eur J Epidemiol, 2009. 24(5): p. 217-24

    work page 2009

  26. [26]

    Yamamoto, and T

    Suzuki, E., E. Yamamoto, and T. Tsuda, Identification of operating mediation and mechanism in the sufficient-component cause framework. Eur J Epidemiol, 2011. 26(5): p. 347-57

    work page 2011

  27. [27]

    VanderWeele, T.J. and T.S. Richardson, General Theory for Interactions in Sufficient Cause Models with Dichotomous Exposures. Ann Stat, 2012. 40(4): p. 2128-2161

    work page 2012

  28. [28]

    VanderWeele, T.J. and J.M. Robins, The identification of synergism in the sufficient- component-cause framework. Epidemiology, 2007. 18(3): p. 329-39

    work page 2007

  29. [29]

    The International Journal of Biostatistics 2011

    Sjolander, A., Reaction to Pearl's Critique of Principal Stratification. The International Journal of Biostatistics 2011. 7(1): p. 22

    work page 2011

  30. [30]

    BMC Med Res Methodol, 2013

    Suzuki, E., et al., A counterfactual approach to bias and effect modification in terms of response types. BMC Med Res Methodol, 2013. 13: p. 101

    work page 2013

  31. [31]

    Suzuki, E. and T.J. VanderWeele, Mechanisms and uncertainty in randomized controlled trials: A commentary on Deaton and Cartwright. Soc Sci Med, 2018. 210: p. 83-85

    work page 2018

  32. [32]

    Eur J Epidemiol, 2008

    Hafeman, D.M., A sufficient cause based approach to the assessment of mediation. Eur J Epidemiol, 2008. 23(11): p. 711-21

    work page 2008

  33. [33]

    Rothman, K.J. and S. Greenland, Causation and causal inference in epidemiology. Am J Public Health, 2005. 95 Suppl 1: p. S144-50

    work page 2005

  34. [34]

    Kezios, K. and E. Hayes -Larson, Sufficient component cause simulations: an underutilized epidemiologic teaching tool. Front Epidemiol, 2023. 3

    work page 2023

  35. [35]

    Gatto, and U.B

    Schwartz, S., N.M. Gatto, and U.B. Campbell, Extending the sufficient component cause model to describe the Stable Unit Treatment Value Assumption (SUTVA). Epidemiol Perspect Innov, 2012. 9: p. 3

    work page 2012

  36. [36]

    Int J Biostat, 2011

    Vanderweele, T.J., Principal stratification--uses and limitations. Int J Biostat, 2011. 7(1)

    work page 2011

  37. [37]

    Epidemiology,

    Lee, W.C., Completion potentials of sufficient component causes. Epidemiology,

    work page

  38. [38]

    always survivors

    23(3): p. 446-53. Appendix Derivations used for construction of Table 2 Table S1 Derivations used for construction of Table 2 As mentioned in the main text, the SACE is written as a weighted average of SACE𝑢. This can be written using the notation in Table 2, as below: 𝐸[𝑌1 − 𝑌0|(𝑆1 = 1, 𝑆0 = 1), 𝑈 = 1] 𝑃((𝑆1 = 1, 𝑆0 = 1), 𝑈 = 1) 𝑃((𝑆1 = 1, 𝑆0 = 1)) + 𝐸[𝑌...

    work page