Visible signs of disease can evoke stigma while stigma contributes to depression and mental illness, sometimes manifesting as somatic symptoms. We assessed these hypotheses among Ebola virus disease (EVD) survivors, some of whom experienced clinical sequelae. Ebola virus disease survivors in Liberia were enrolled in an observational cohort study starting in June 2015 with visits every 6 months. At baseline and 18 months later, a seven-item index of EVD-related stigma was administered. Clinical findings (self-reported symptoms and abnormal findings) were obtained at each visit. We applied the generalized estimating equation method to assess the bidirectional concurrent and lagged associations between clinical findings and stigma, adjusting for age, gender, educational level, referral to medical care, and HIV serostatus as confounders. When assessing the contribution of stigma to later clinical findings, we restricted clinical findings to five that were also considered somatic symptoms. Data were obtained from 859 EVD survivors. In concurrent longitudinal analyses, each additional clinical finding increased the adjusted odds of stigma by 18% (95% CI: 1.11, 1.25), particularly palpitations, muscle pain, joint pain, urinary frequency, and memory loss. In lagged associations, memory loss (adjusted odds ratio [AOR]: 4.6; 95% CI: 1.73, 12.36) and anorexia (AOR: 4.17; 95% CI: 1.82, 9.53) were associated with later stigma, but stigma was not significantly associated with later clinical findings. Stigma was associated with select symptoms, not abnormal objective findings. Lagged associations between symptoms and later stigma substantiate the possibility of a pathway related to visible symptoms identified by community members and leading to fear of contagion.

Many different methods for evaluating diagnostic test results in the absence of a gold standard have been proposed. In this paper, we discuss how one common method, a maximum likelihood estimate for a latent class model found via the Expectation‐Maximization (EM) algorithm can be applied to longitudinal data where test sensitivity changes over time. We also propose two simplified and nonparametric methods which use data‐based indicator variables for disease status and compare their accuracy to the maximum likelihood estimation (MLE) results. We find that with high specificity tests, the performance of simpler approximations may be just as high as the MLE.