Reducing attrition through a vital status tracing methodology in Swiss spinal cord injured patients

Abstract

Introduction Attrition, or loss to follow-up (LTFU), in longitudinal studies can result in selection bias and hinder study validity. Previous researches on attrition in chronic disease populations have suggested attrition to be associated with an increased risk of mortality. Spinal cord injuries (SCI) are a chronic disease with evidenced reductions in life expectancy and increased premature mortality. If an individual's risk of adverse outcomes is associated with irregular or discontinued care in a SCI-specialized rehabilitation center, loss to clinic would be an additional concern for the rehabilitation setting. Given the potential impact of attrition for this population, a comprehensive tracing strategy was undertaken to ascertain and update the vital status (VS) of persons included in the Swiss Spinal Cord Injury (SwiSCI) cohort study. The objective of this study is to therefore outline the tracing methodology used to ascertain vital status, and to assess risk factors for attrition and loss to clinic, and the potential impact of loss to clinic on longevity outcomes. Methods Vital status of persons with SCI was updated through medical records of SwiSCI-covered specialized rehabilitation centers and, where-needed, tracing through Swiss municipalities. Lost to clinic (LTC) and lost to follow-up (LTFU) were defined as having no available VS information 18 months before the population censoring date (September 30, 2011), with the latter involving the additional tracing through one or more municipalities. For a proportion of cases, municipalities were contacted without first checking medical records due to practicality issues (e.g., historical, paper-based records). Risk factors for LTC were assessed using a flexible parametric survival model, with time since injury until LTC (yes/no) as the underlying timescale. In a secondary analysis, hazard ratios (HRs) for risk of mortality (dead/alive) among those LTC compared to not LTC were estimated using a flexible parametric survival model, whereas follow-up time was split at date of LTC, if applicable. Risk factors for LTFU (yes/no) were assessed using logistic regression, which was conditional on being LTC. Results In total, 3270 individuals were included in the vital status update. In total, 1041 individuals were identified as LTC and required additional tracing through municipalities; among those with an identified VS (N = 889), 336 had died by the cutoff date. Among individuals with a traumatic SCI (TSCI), risk factors for LTC included: incomplete lesions, older age at injury, and rehabilitation center. Individuals with a non-traumatic SCI (NTSCI) had a higher risk of going LTC is they were female and older at time of NTSCI; contrary to individuals with TSCI, lesion characteristics did not have an effect on risk of LTC. Additionally, preliminary results suggest that individuals LTC who have incurred a TSCI have a nearly five-fold increase in risk of premature mortality compared to those not LTC (HR = 4.83; 95% confidence interval [CI] = 2.87–8.12). There was no evidence of a risk differential among individuals with an NTSCI (HR = 1.29; 95% CI = 0.91–1.84). Finally, both time since LTC and age at SCI were associated with LTFU, while the probability of going LTFU declined with age. Of the 1′368 cases that were traced directly through municipalities - and thereby excluded from analyses - VS was ascertained for 75.8% of cases, of which 546 (52.7%) had died. Conclusion The impact of LTC status on risk of premature mortality could be indicative of the value of continued specialized care in the years following initial trauma and beyond first rehabilitation. Additional research is thus needed that can verify the true impact of continued specialized care on mortality outcomes. Furthermore, this study confirms the likelihood of attrition to cause selection bias in the evaluation of mortality and life expectancy. Results from this study can help facilitate the correction of mortality estimates in future analyses.