Factors influencing estimated COVID-19 vaccine effectiveness: outcomes affected and type of influence
| Factors | Outcomes affected* | Influence on effectiveness estimate† |
| Pre-existing immunity | ||
| Same in V and UNV | I, H, D | Genuine |
| Different in V and UNV | I, H, D | Spurious (selection bias) |
| Vaccination misclassification | I, H, D | Spurious (misclassification) |
| Exposure difference | ||
| Induced by perceived vaccine protection | I, H, D | Genuine |
| Pre-existing, carried forward | I, H, D | Spurious (selection bias) |
| Testing | ||
| Typical diagnosis bias | I, H, D | Spurious (selection bias, misclassification) |
| Affecting treatment | H, D | Genuine |
| Disease risk factor confounding | H, D | Spurious (confounding bias) |
| Hospital admission decision | ||
| Induced by perceived vaccine protection | H, (D)‡ | Genuine |
| Other reasons | H, (D)‡ | Spurious (selection bias, confounding) |
| Treatment use difference | ||
| Induced by perceived vaccine protection | D, (H)§ | Genuine |
| Other reasons | D, (H)§ | Spurious (selection bias, confounding) |
| Death attribution | D | Spurious (misclassification) |
*The outcomes considered here are infection (I), hospitalisation (H) and death (D).
†There are two major types of influence: genuine impact represents effect modification (ie, the vaccine effectiveness is genuinely different in different groups, settings, etc) and conversely, spurious impact means bias (ie, the vaccine effectiveness is different from what is being estimated).
‡May affect death outcome if hospital admission affects the risk of death (not necessarily so).
§For treatments that are applied before hospitalisation and may affect the need for hospitalisation.
UNV, unvaccinated; V, vaccinated.