Table 2.

Bias aspects of studies included in the meta-analyses

First author, year	Misclassification and measurement error of outcome		Control of confounding			Selection bias/loss to follow-up		Risk of bias*
			Socioeconomic control	Time varying exposure control				A	B	C	D	Overall
Active case ascertainment studies
Oudin, 201616	Mostly active: in-person assessment supplemented with MR (some MR only)		Individual level adjustment	Not time varying exposure		Full follow-up		Some	Low	Low	Some	Some
Astrom 202120	Mostly active: in-person assessment supplemented with MR (some MR only)		Individual level adjustment	Not time varying exposure		Full follow-up		Some	Low	Low	Some	Some
Wang, 202223	Active: in-person or telephone screening followed by in-person assessment		Individual and area level adjustment; Adjustment for potential mediators	Not time varying exposures		Weighting to address loss to follow-up		High*	Low	Some	Low	High*
Grande, 202025	Active: in-person assessment supplemented with death and medical records		Individual level adjustment; Adjustment for potential mediators	Time varying exposure with adjustment for time trend		6-11% loss to follow-up		High*	Low	Some	Low	High*
Paul, 202026	Active: in-person screening, with neuropsychological exam follow-up reviewed by neurologist and neuropsychologist review		Individual and area level adjustment	Not time varying exposures		Weighting to address loss to follow-up		Some	Low	Some	Low	Some
Mortamais 202129	Active: 3 phase in-person assessment, review by geriatric specialist		Individual level adjustment	Not time varying exposure		Weighting to address loss to follow-up		High	Low	Some	Low	High
Semmens, 202130	Active; screening followed by neuropsychological battery, neurological evaluation and adjudication		Individual and area level adjustment	Not time varying exposure		Approaches to limit selective attrition in study design		Some	Low	Some	Low	Some
Shaffer 202131	Active: in-person assessments, follow-up physical, neuropsychological evaluations reviewed by consensus		Individual and area level adjustment	Time varying exposure with adjustment for time trends		14% loss to follow-up		Some	Low	Some	Low	Some
Sullivan, 202132	Active: in-person Clinical Dementia Rating assessment by trained interviewers (case=rating >1)		Individual level adjustment	Time varying exposure, no adjustment for time trends		No information on loss to follow-up		High	Low	High*	Low	High
Passive case ascertainment studies
Chen, 201737	Passive: ICD codes and prescriptions		Area level adjustment	Time varying exposures; reported no difference in results with time trend adjustment		Likely low; province wide data, required residence >5 years		High	High	Low	High*	High
Carey, 201839	Passive: Primary care record Quality and Outcomes Framework Read Codes, and death ICD codes		Area level adjustment	Not time varying exposures		Censored if GP withdrew from CPRD; patient loss to follow-up not known		Some	High	Low	High*	High
Cerza, 201943	Passive: ICD codes		Area level adjustment	Not time varying exposure		Loss to follow-up discussed		High	High	Low	High*	High
Smargiassi, 202052	Passive: ICD codes and prescriptions		Area level adjustment	Time varying exposure with time trend adjustment		Likely low; province wide data, required residence >4 years		High	High	Low	High*	High
Ran, 202156	Passive: ICD codes		Individual level adjustment	Not time varying exposure		Likely none; Hong Kong wide data		Some	High	Low	High*	High
Shi, 202159	Passive: ICD codes		Individual (Medicaid eligibility) and area level adjustment	Time varying exposure with time trend adjustment		Likely low; all ≥65 years; nationwide data		Some	High	Low	High*	High
Parra, 202262	Passive: ICD codes		Individual level adjustment	Not time varying exposure		Censoring for loss to follow-up		Some	High	Low	High*	High

CPRD=Clinical Practice Research Datalink; GP=general practitioner; ICD=International Classification of Diseases MR=medical records.

^*Indicates that the likely direction of bias would be towards the null and no other bias is greater than some. Risk of bias domains: A=Confounding; B=Post-exposure intervention; C=Missing data; D=Measurement of the outcome. All studies were rated some risk of bias in the domains of “Measurement of the exposure” and “Selection of reported results,” and low risk of bias in the domain of “Selection of participants.”