Process of estimation of the AF | Equations | Key bias sources | Bias model parameter for MCSA (suffices 0, t indicate chosen and unknown true values, respectively) |
(1) Point estimate of the numbers exposed | n0 = pwk(E)*nemployed | Biases acting on point estimate of numbers exposed (n0) 1. Inadequate data for numbers employed (nemployed) 2. Unknown proportion exposed [pwk(E)] 3. High/low exposure level split inaccurate | Cn=n0/nt |
(2) Number ever exposed accounting for turnover and latency* | Ne(REP)=n0 + {n0 * TO * t} | 4. Unknown or inaccurate latency and thus the length of the REP (t) | Cl=t0/tt |
5. Inaccurate staff turnover (TO) | Ct= TO0/TOt | ||
(3) Proportion of the population exposed | p(E) = Ne(REP) / Np(REP) | ||
(4) AF: Levin’s equation | AF = p(E)*(RR–1)/{1+p(E)*(RR–1)}, extended to AF = {Σjp(Ej)(RRj-1)}/{1+{Σj p(Ej)(RR-1)} summed over j separate exposure level/industry sector/time period combinations | 6. Random error affecting RR | |
7. Bias affecting RR chosen from an inappropriate published study regarding distribution of confounders | Cr= RR0/RRt | ||
8. Bias affecting RR chosen from a published study that is inaccurately matched to GB exposure situation, or is subject to exposure misclassification or publication bias | Cm= RR0/RRt | ||
9. Bias from using adjusted estimate of RR (RRa) incorrectly instead of unadjusted (RRu) in Levin’s equation16 | Cu= RRa/RRu | ||
(5) Estimating attributable numbers | AN=AF*∑jDj summed over the j age-groups of the REP exposed population in the target year | 10. Bias due to not estimating age-specific AFs |
*This equation is modified when estimating AF to account for recruitment and retirement ages and survival of exposed workers by the target year.6
AF, attributable fraction; AN, attributable numbers; D, deaths /registrations; MCSA, Monte Carlo sensitivity analysis; n0,numbers exposed at a particular year in the REP; nemployed, total numbers employed; pwk(E), proportion in the workplace exposed; TO, staff turnover per year; t, latent period, ie, number of years in the REP; Ne(REP), numbers ever exposed during the REP; Np(REP), numbers ever of working age during the REP; p(E), proportion of the population exposed; REP, risk exposure period; RR, relative risk.