โ† Wiki Index / Public Health & Global Health / Depletion-of-Susceptibles Bias
๐Ÿฅ Public Health & Global Health

Depletion-of-Susceptibles Bias

Depletion-of-Susceptibles Bias is a longitudinal selection bias in pharmacoepidemiology and vaccine studies where differential attrition of high-risk individuals distorts period-specific hazard ratios over time.

Depletion-of-Susceptibles Bias (also termed survivor-cohort selection bias) is a subtle longitudinal selection bias in epidemiology and survival analysis that arises when individuals within a cohort differ in baseline susceptibility or frailty, and the more susceptible individuals experience the outcomeโ€”and thus exit the risk setโ€”faster in one exposure arm than in the other. As follow-up progresses, the remaining "at-risk" pool in the higher-risk arm becomes disproportionately enriched with resilient or low-risk survivors, causing period-specific hazard ratios to drift artificially over time even when individual treatment efficacy remains constant.

Mechanism in Pharmacoepidemiology and Vaccine Waning

Depletion-of-susceptibles bias appears prominently in two distinct clinical settings:

Drug Safety and Prevalent-User Studies

Suppose a medication (such as hormone replacement therapy or an NSAID) elevates the immediate hazard of venous thromboembolism among genetically or clinically susceptible users. Early after treatment initiation, susceptible users suffer events and discontinue the drug or are censored. If researchers compare prevalent long-term users against non-users, the remaining treated cohort consists largely of non-susceptible survivors, making the drug appear falsely safe or even protective at later intervals.

Apparent Waning of Vaccine Effectiveness

In infectious-disease cohorts, suppose a vaccine reduces infection risk uniformly by 70 percent across individuals who vary in social contact rate or biological susceptibility. During an epidemic wave, high-exposure unvaccinated individuals get infected quickly and acquire natural immunity, depleting the unvaccinated susceptible pool faster than the vaccinated pool. Consequently, in later months, the unvaccinated control group has a lower average baseline risk than the vaccinated groupโ€”creating an illusory decline in measured vaccine effectiveness (VE) that mimics biological waning immunity even when true individual protection has not decayed.

Clinical ContextFaster Depletion ArmResulting Distortion at Late Follow-Up
Adverse Drug EventTreated arm (early events)Hazard ratio drops toward or below 1.0 (false safety)
Epidemic / Vaccine WaningUnvaccinated arm (faster natural infection)Apparent VE drops over time (illusory waning)
Homogeneous Risk PoolEqual baseline susceptibilityZero depletion-of-susceptibles bias

Quantitative Remedies

Methodologists mitigate depletion-of-susceptibles bias using new-user (incident-user) designs, estimating cumulative risk differences or marginal survival curves rather than conditional period-specific hazard ratios, stratifying by prior infection or contact history, and using time-dependent propensity scores or g-methods to account for differential survivor selection.

Distinguishing True Waning from Statistical Attrition

During the rollout of SARS-CoV-2 and seasonal influenza vaccines, distinguishing genuine biological waning of neutralizing antibodies from depletion-of-susceptibles bias became a central methodological priority. As Lipsitch et al. (2019) demonstrated via mathematical simulation, even a "leaky" vaccine conferring constant 70 percent per-exposure protection exhibits falling period-specific hazard ratios over an epidemic wave whenever unvaccinated high-contact individuals acquire natural immunity first. In contrast, an "all-or-nothing" vaccine (protecting 70 percent of recipients completely and 30 percent not at all) shows even stronger depletion dynamics in the vaccinated unprotected subgroup unless baseline exposure heterogeneity is controlled.

Practical Remedies in Cohort Analysis

Modern vaccine-waning and pharmacoepidemiology protocols counter depletion-of-susceptibles bias by restricting follow-up to infection-naive or incident-user cohorts confirmed via serology, aligning calendar time and baseline risk exposure across treated and comparator arms (target-trial emulation), and evaluating cumulative incidence curves or restricted mean survival differences rather than relying solely on late-period conditional hazard ratios.

Gamma-Frailty Attenuation and Conditional Hazard Drift

In survival analysis, depletion-of-susceptibles bias can be formalized with a multiplicative frailty model. Suppose each person carries a latent, time-invariant susceptibility multiplier Z with mean 1 and variance ฮฝ > 0 following a Gamma distribution, and true conditional hazard h(t | Z, X) = Z ยท h0(t) ยท exp(ฮฒX) for binary treatment X โˆˆ {0, 1} with constant biological hazard ratio HRtrue = exp(ฮฒ). Letting H0(t) denote the cumulative baseline hazard up to time t, conjugate updating implies that among individuals who remain event-free at time t (T โ‰ฅ t), the expected residual frailty is E[Z | T โ‰ฅ t, X] = 1 / [1 + ฮฝ ยท H0(t) ยท exp(ฮฒX)].

At t = 0 (H0(0) = 0), both groups share mean frailty 1, so the marginal hazard ratio equals HRtrue. Whenever the intervention is protective (exp(ฮฒ) < 1), the control arm (X = 0) sheds high-frailty members faster than the treated arm (X = 1), raising E[Z | T โ‰ฅ t, X = 1] relative to E[Z | T โ‰ฅ t, X = 0]. Consequently, the population-averaged period hazard ratio becomes HRmarg(t) = exp(ฮฒ) ยท [1 + ฮฝ ยท H0(t)] / [1 + ฮฝ ยท H0(t) ยท exp(ฮฒ)]. As cumulative incidence rises (H0(t) โ†’ โˆž), HRmarg(t) converges smoothly toward 1 even though the within-person biological effect exp(ฮฒ) has not changed. In discrete-frailty or high-attack-rate cohorts where highly exposed untreated individuals exhaust rapidly, period-specific hazard ratios can even exceed 1.0 late in follow-up, making cumulative incidence curves, target-trial emulation, and shared-frailty sensitivity checks essential before concluding that drug or vaccine efficacy wanes biologically.

Frequently Asked Questions

Does depletion-of-susceptibles bias occur if everyone has the exact same risk?

No. The bias requires unobserved heterogeneity in baseline susceptibility or exposure intensity across individuals in the cohort; when ฮฝ = 0, residual risk sets remain exchangeable over time and HRmarg(t) equals HRtrue.

Why do hazard ratios suffer from this bias more than cumulative incidence curves?

A hazard ratio conditions on having survived event-free up to time t, which breaks initial randomization or baseline exchangeability whenever one arm loses high-risk individuals faster than the other. Cumulative incidence curves condition solely on baseline cohort entry at t = 0.

References

  1. Lipsitch, M., Goldstein, E., Ray, G. T., & Fireman, B. (2019). Depletion-of-susceptibles bias in influenza vaccine waning studies: how to ensure robust results. Epidemiology and Infection, 147, e306. https://doi.org/10.1017/s0950268819001961
  2. Renoux, C., Dell'Aniello, S., Brenner, B., & Suissa, S. (2017). Bias from depletion of susceptibles: the example of hormone replacement therapy and the risk of venous thromboembolism. Pharmacoepidemiology and Drug Safety, 26(5), 554โ€“560. https://doi.org/10.1002/pds.4197
  3. Kahn, R., & Feikin, D. R. (2023). Examining Bias From Differential Depletion of Susceptibles in Vaccine Effectiveness Estimates in Settings of Waning. American Journal of Epidemiology, 192(11). https://doi.org/10.1093/aje/kwad191
๐Ÿงฌ Support Open Science
Explore interactive science games on LowKey Games or browse scientific apparel & accessories in the CRV Science Shop.
๐Ÿ’ก Spot an inaccuracy or have primary literature to expand this article?
Submit a citation or revision note using your shared CRV Science / LowKey Games profile ID.
Suggest Revision