Survival analysis under imperfect record linkage using historic census data

Our plain-English summary of Survival analysis under imperfect record linkage using historic census data, by A.K. Marks-Anglin, F.K. Barg, M. Ross, et al (BMC Medical Research Methodology, 2024).

The study

• Authors: A.K. Marks-Anglin, F.K. Barg, M. Ross, D.J. Wiebe, W.T. Hwang
• Journal: BMC Medical Research Methodology
• Published: 2024;24(1):67
• Cited by: 2 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Background: Advancements in linking publicly available census records with vital and administrative records have enabled novel investigations in epidemiology and social history. However, in the absence of unique identifiers, the linkage of the records may be uncertain or only be successful for a subset of the census cohort, resulting in missing data. For survival analysis, differential ascertainment of event times can impact inference on risk associations and median survival.

Methods: We modify some existing approaches that are commonly used to handle missing survival times to accommodate this imperfect linkage situation including complete case analysis, censoring, weighting, and several multiple imputation methods. We then conduct simulation studies to compare the performance of the proposed approaches in estimating the associations of a risk factor or exposure in terms of hazard ratio (HR) and median survival times in the presence of missing survival times. The effects of different missing data mechanisms and exposure-survival associations on their performance are also explored. The approaches are applied to a historic cohort of residents in Ambler, PA, established using the 1930 US census, from which only 2,440 out of 4,514 individuals (54%) had death records retrievable from publicly available data sources and death certificates. Using this cohort, we examine the effects of occupational and paraoccupational asbestos exposure on survival and disparities in mortality by race and gender.

Results: We show that imputation based on conditional survival results in less bias and greater efficiency relative to a complete case analysis when estimating log-hazard ratios and median survival times. When the approaches are applied to the Ambler cohort, we find a significant association between occupational exposure and mortality, particularly among black individuals and males, but not between paraoccupational exposure and mortality.

Discussion: This investigation illustrates the strengths and weaknesses of different imputation methods for missing survival times due to imperfect linkage of the administrative or registry data. The performance of the methods may depend on the missingness process as well as the parameter being estimated and models of interest, and such factors should be considered when choosing the methods to address the missing event times.

© 2024. The Author(s).

In plain English

The authors compared statistical methods for handling missing death dates when census records cannot be linked to death records. They tested them in simulations and in Ambler, Pennsylvania, 1930-census residents (4,514 in the cohort, 4,507 analyzed). Death dates were found for 2,440 people (54%). Death dates were found more often for U.S.-born, older, male and white residents. Occupational exposure meant a listed workplace such as an asbestos or shingles plant. Paraoccupational exposure meant sharing an address with an occupationally exposed person or, lacking a listed house number, being a listed relative. The outcome was death from any cause. Simulations found conditional-survival imputation most consistently reduced bias. Occupational exposure was linked to higher mortality in some analyses (inverse probability weighting hazard ratio 1.16, 95% CI 1.004 to 1.34). The link was strongest in black residents: 1.50 (1.06 to 2.12), same method. In Cox models, paraoccupational exposure showed no significant link to mortality except with restricted-mean imputation in black residents. Accelerated failure-time models with inverse probability weighting gave black residents a paraoccupational event-time ratio of 0.94 (0.89 to 0.996). The authors assumed recorded variables explained which deaths were found.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Statistical methods study: computer simulations comparing missing-data methods (complete case, inverse probability weighting, censoring, two multiple-imputation methods), then applied to a historical census-based cohort
• Who was studied: Cohort of 4,514 residents of Ambler, PA listed in the 1930 U.S. census (described as adult residents). The analysis included 4,507 with complete covariate information.
• Exposure: Occupational exposure: listed place of work was asbestos, shingles plant, shingle mill, chemical plant/works, or mill (10.5% of the cohort). Paraoccupational exposure: same residential address as an occupationally exposed person, or, for those without a listed house number, a listed familial relationship to one (36.2%). No fiber type or dose was given.
• Main results:
  • The outcome was overall mortality. Death dates were identified for 2,440 of the cohort (54%).
  • Death dates were more likely to be found for U.S.-born, older, male and white residents (ascertainment bias).
  • In simulations, multiple imputation based on conditional survival (MICS) most consistently reduced bias across settings.
  • Whole analyzed cohort (n = 4507), occupational exposure, Cox model hazard ratios: IPW 1.16 (95% CI 1.004, 1.34); complete cases 1.18 (1.02, 1.36); MICS 1.14 (0.999, 1.30).
  • Black residents (n = 560), occupational exposure hazard ratios: complete cases 1.59 (1.10, 2.30); IPW 1.50 (1.06, 2.12); MICS 1.31 (1.002, 1.70). Male residents (n = 2282): IPW 1.17 (1.01, 1.36).
  • In the Cox model, differences by paraoccupational exposure were not statistically significant, except with MIRM among black residents: hazard ratio 1.57 (1.16, 2.11) with tau = 80, and 1.53 (1.11, 2.12) with tau = 110.
  • In the Weibull accelerated failure-time model (Table 8), the event-time ratios for paraoccupational exposure among black residents were: IPW 0.94 (0.89, 0.996); MIRM tau = 80 0.95 (0.92, 0.97); MIRM tau = 110 0.95 (0.92, 0.98). The IPW event-time ratio for occupational exposure in this group was 0.91 (0.84, 0.99).
  • Median survival was reported as lower among occupationally or paraoccupationally exposed individuals than among unexposed ones. For example, among complete cases in the whole cohort it was 69.90 years for the occupationally exposed versus 72.33 years for the unexposed.
• The authors' conclusion: The authors concluded that how well each missing-data method works may depend on the missingness process, the parameter being estimated and the model used, and that MICS most consistently reduced bias in their simulations.
• Limitations the authors state:
  • They assumed the survival and imputation models were correctly specified.
  • They did not consider interactions between covariates and exposure in the missingness or survival mechanisms.
  • They did not vary the level of missingness or censoring in the simulations.
  • In the Ambler application, they assumed the observed variables were enough to account for differential ascertainment of deaths.
  • They assumed non-informative censoring, which may not hold because life expectancy and record-keeping both improved over time.

Topics

Medical terms in this study, explained in our Lexicon: Epidemiology, Median Survival, Hazard Ratio, Death Certificate, Asbestos Exposure, Occupational Exposure.

Related studies

• Mapping cancer risk in southwestern Ontario with changing census boundaries — P. Nguyen, P.E. Brown, et al. (2012)
• Management and Survival of Pleural Mesothelioma: A Record Linkage Study — G. Carioli, M. Bonifazi, et al. (2018)
• Ascertainment Bias in a Historic Cohort Study of Residents in an Asbestos Manufacturing Community — J.D. Wortzel, D.J. Wiebe, et al. (2021)
• Hospital records as a data source for occupational disease surveillance: a feasibility study — J. Balmes, D. Rempel, et al. (1992)
• Is pleural infection associated with longer survival in mesothelioma? A population-based cohort study using data from Hospital Episode Statistics — A.C. Bibby, D. de Fonseka, et al. (2019)
• Medically induced euthyroid hypothyroxinemia may extend survival in compassionate need cancer patients: an observational study — A. Hercbergs, R.E. Johnson, et al. (2015)

Browse return to medical research.