Table 1.
Potential biases that can affect the estimation of CFR (and thereby also the comparison of CFR across groups).
Fig 1.
Illustration of delayed reporting bias in an exponentially growing epidemic.
In an ongoing epidemic, there will typically be a delay between the reporting of a case and the reporting of the death of that case, if the infected person dies. Thus, at any moment, there will be some cases reported who will die of the infection but who have not yet died, or whose deaths have not yet been reported. Simple division of the number of deaths reported by week w (green), by the number of cases reported by week w (blue) will underestimate the CFR because the numerator does not include all those cases in the denominator who will eventually die. With a reporting delay of 3 weeks for deaths compared to cases, the reported deaths curve will be shifted 3 weeks to the right, relative to the curve of the total number of cases reported by week w who will die (red). If the epidemic doubling time is 2 weeks, as shown here, the underestimate of CFR will be by a factor of about 23/2 ≈ 2.8, with the exponent being the number of epidemic doubling times that pass between case reporting and death reporting. In reality, there will be a distribution of reporting delays rather than a fixed delay, making this a heuristic rather than exact approach. The problem is ameliorated in an epidemic that grows more slowly or less than exponentially. For more details, see references in Table 1.
Table 2.
Potential biases that can affect the comparison of CFR across groups (relative CFR), using the example of comparing the CFR among hospitalized and non-hospitalized persons to assess the relative CFR for hospitalization.
Table 3.
Effect of selection bias on estimates of relative CFR on the risk ratio (RR) and odds ratio (OR) scale.
Table 4.
Effect of selection bias on estimates of relative CFR on the risk ratio (RR) and odds ratio (OR) scale.
Table 5.
Effect of selection bias on estimates of relative CFR on the risk ratio (RR) odds ratio (OR) and risk difference (RD) scales.