1918 Flu Study: Healthy, Frail Died at Same Rates

A paper authored by anthropologists Drs. Amanda Wissler and Sharon DeWitte, published October 2023 in The Proceedings of the National Academy of Sciences, uses evidence from skeletal samples to challenge the perception that the 1918 influenza pandemic killed indiscriminately regardless of age or well-being. The skeletal samples, sourced from the Cleveland Museum of Natural History’s Hamann-Todd collection, included 81 individuals who died between September 1918 and March 1919, as well as 288 in a control group who died after 1910 but before the pandemic.

Influenza has wreaked havoc on human populations for millennia, with the first written reference to an influenza-like epidemic dating back to the fifth century BCE in Hippocrates’s Epidemics.

Normally, selective mortality places frailer or disadvantaged individuals at greater risk of early death than healthy individuals. Prior research, according to the paper, has shown that during the average “catastrophic death event” like an epidemic, this selective force is often reduced but not completely removed, as we saw during the COVID-19 pandemic.

Countless contemporary reports from the 1918 pandemic, which infected approximately one-third and killed between 1.3–3% of the world population, however, describe victims as young, healthy adults who died at the same rates as frailer groups like children and the elderly, suggesting the complete elimination of selective mortality.

In the century since, different theories have been proposed: older adults’ previous exposure to a different strain of the same virus in a previous and less serious epidemic, the “cytokine storm”—essentially the overreaction of a strong immune system, and misinformation in the time period.

Wissler and DeWitte argue that no concrete scientific data currently exists to support these anecdotes and that sources like vital statistics, census data, and life insurance records are not adequate in characterizing victims as “healthy,” providing little information on comorbidities or previous health conditions.

In their study, they determine frailty status using the state of periosteal lesions in the skeletons’ tibiae as an indicator—active lesions, according to the paper, have been previously correlated with greater frailty than mixed-status lesions (both active and healed tissue) or healed lesions (with “smooth, remodeled bone”). 200 control skeletons and 48 influenza-era skeletons were categorized in this way; the rest of the tibiae examined in the study did not have any lesions and thus were discarded from analysis.

With each group—control and flu—having three lesion statuses, the authors used Kaplan-Meier curves to analyze the estimated probability of survival as well as Cox proportional hazard models for a general impact of different lesion statuses on the chances of early death.

In Kaplan-Meier curves, each drop is a death, and the overall shape of the graph should approach the true survival probability function. Wissler and DeWitte calculated no statistically significant difference between the mixed and healed lesions lines, which are green and red on the graphs, respectively. In the flu group, the median survival time for active-lesion individuals was 2.5 years fewer than healed-lesion and nine years fewer than mixed-lesion individuals (36, 38.5, and 45, respectively); this result opposes the null hypothesis that selective mortality did not exist during the pandemic. In comparison, the control group’s median survival times were 39, 45, and 46 years.

The paper’s calculated hazard ratios (HRs) via Cox proportional analysis, which assumes the same hazard ratio to be constant, and thus gives an average statement, indicates that within the individuals examined in the study, those with active lesions were 2.76 times as likely to die as mixed-lesion individuals. P-value, or the probability of the same or more extreme outcome, was calculated to be statistically significant at 0.065 (or 6.5%), thus supporting that this result is unlikely to be a coincidence; the statistical significance level was set at 0.1. The p-value for the hazard ratio between mixed- and healed-lesion individuals was 0.21 and thus is not statistically significant. Overall, Wissler and DeWitte rejected their null hypothesis and stated that “healthy individuals were not equally likely to die as frail individuals during the 1918 flu.”

Although this study and paper used lesion status to estimate frailty, the authors acknowledge the incomplete picture this gives—individuals with no lesions were eliminated and could either be people who died before lesions could properly form or people whose bodies were strong enough not to develop lesions.

The small sample size that remained did not help, since it led to a large confidence interval for results and was too small to be able to do many statistical analyses. Convention sets p-value significance at 0.05, and perhaps the authors used 0.1 to allow for a greater margin of error due to this factor.

Additionally, the skeletons used were all unclaimed individuals in Cleveland, which must be taken into account in generalizing the study. The authors state the unclaimed status may indicate lower socioeconomic status.

In analyzing concrete remains, Wissler and DeWitte’s study as well as future similar bioarchaeology may put forth new information to finally help bring to light what exactly happened in a deadly pandemic of which the details, in many ways, are shrouded in uncertainty and overshadowed by World War I.