Table 2.

Sources of bias, analytical approach and limitations in context using historic controls as the comparison group

Possible source of bias	Problem	Analytical approach in the analysis plan to reduce bias	Implications
Survival bias	Definition of a comparable starting point for follow-up for both groups: PCR confirmation of diagnosis may not be received on the same day the sample was taken; Upon confirmation of diagnosis, plasma treatment could be initiated on that day or up to 2 days later; A comparable date for starting plasma treatment could not be defined for historic control patients who were not assessed for eligibility to receive convalescent plasma.	Exclude deaths occurring up to and including the second day after confirmation of Ebola virus disease diagnosis.	Patients in both groups have a comparable starting point of the third calendar day after diagnosis (Figure 2).
			Unbiased comparison of convalescent plasma and historic patients (with respect to follow-up time). Case fatality rate underestimated in both groups as early deaths excluded (Figure 2), but “intention-to-treat” analysis also performed.
	With a starting point of day 3, survival status 14 days post-diagnosis may not allow a full 14 days of follow-up after the onset of treatment in the convalescent plasma group.	The primary outcome was measured as survival status 16 days after diagnosis for both plasma-treated and historic patients, as plasma administration started up to 2 days after diagnosis.	Analysis of the primary endpoint becomes mortality between the 3rd and 16th days post-diagnosis for a harmonized follow-up period. Few deaths from Ebola virus disease were expected after day 16 in either group.
	Patients were discharged once confirmed cured. In the trial, patients were contacted by phone to confirm survival status up to day 30, but this was not done for the historic patients.	Assume historic patients discharged confirmed cured before day 16 were alive atday 16.	It is possible that deaths between days 16 and 30 could be missed for historic patients, but these would likely be due to concomitant conditions or possible sequelae and occur very rarely. One death occurred between days 16 and 30 in a plasma-treated patient after being discharged cured of Ebola disease due to another concomitant condition. Comparison of mortality rates was confined to day 16 only.
		Patients still hospitalized on day 16 are counted as survivors in the primary analysis.
Measurement bias and residual confounding	Accurate data capture was expected for objective factors such as age, sex and PCR cycle threshold as an indicator of viral load.	A priori adjustment for more objective measures of prognostic factors of age and PCR cycle threshold value at diagnosis, based on available literature.	Clinical symptom data may be subjective and there could be variability in reporting in the two periods or missing data for historic series patients if staff had little time for complete data collection and checking. Opportunities may have missed to detect imbalance between groups.
	Although the same variables of interest were being recorded with similar data collection forms by the same on-site team for historic and trial series, there may have been bias in capture of clinical symptom data over time.	Adjust for baseline factors associated with survival in historic patients and also found to be unbalanced between convalescent plasma and historic patient groups.	Effects of confounding taken into account for measured factors, but limitations remain in case of measurement error or reporting bias.
	Patient symptoms and clinical characteristics associated with survival may be imbalanced between treatment groups through a real effect or as a result of reporting bias.
	Bias toward better survival in trial period, either through better care with additional trial focus on-site or due to lower caseload.		Residual bias may be present.
Treatment bias	Supportive care administration could be variable over time.	Although the trial and historic data are from the same Ebola treatment center, managed by the same organization over time, it is not possible to adequately measure or account for variability in supportive care provision in the analysis.	Residual bias may be present.
	Receipt of additional experimental treatments such as Favipiravir would not allow evaluation of convalescent plasma with supportive care versus supportive care alone.	Exclude patients who received additional experimental treatments.	Smaller sample size but more appropriate analysis group.

PCR: polymerase chain reaction.