Table E3.
Description, justification, and summary results of sensitivity analyses
| Analysis | Description | Justification | Depression |
Anxiety |
||||
|---|---|---|---|---|---|---|---|---|
| No. | Events/PYAR | Adjusted HR∗ (99% CI) | No. | Events/PYAR | Adjusted HR∗ (99% CI) | |||
| Main analysis | 1,980,710 | 134,204/11,290,052 | 1.14 (1.12-1.16) | 2,242,905 | 106,420/12,730,883 | 1.17 (1.14-1.19) | ||
| Sensitivity analysis 1 | Repeating the primary analysis using progressively less-strict definitions of psychiatric diagnoses | To explore potential bias introduced by low sensitivity to detect psychiatric diagnoses in electronic health records, as well as by general practitioner's use of symptom codes, instead of diagnostic codes | ||||||
| (1a) Initially including symptom codes in the definitions of outcomes | 1,980,710 | 211,534/10,970,276 | 1.16 (1.14-1.17) | 2,242,905 | 175,874/12,420,852 | 1.18 (1.16-1.20) | ||
| (1b) Subsequently also adding “nondefinitive” diagnostic codes | 1,980,710 | 227,393/10,908,249 | 1.15 (1.14-1.17) | 2,242,905 | 202,679/12,353,235 | 1.18 (1.16-1.20) | ||
| Sensitivity analysis 2 | Repeating the primary analysis separately for prevalent and incident atopic eczema cases. Stratifying the analysis on the time since the initial diagnosis (0-4 or ≥5 y) | To separate “true prevalent” cases from likely incident atopic eczema cases to explore possible bias due to the choice of a “prevalent” cohort design | ||||||
| (2a) “Incident” cohort (2b) “Prevalent” cohort |
1,431,318 549,392 |
97,372/8,445,494 36,832/2,844,558 |
1.17 (1.15-1.20) 1.05 (1.01-1.09) |
1,646,703 596,202 |
77,545/9,614,471 28,875/3,116,412 |
1.19 (1.16-1.22) 1.10 (1.06-1.15) |
||
| Sensitivity analysis 3 | Repeating the primary analysis including only those who consulted their general practitioner in the year before cohort entry | To explore potential bias due to differential recording of exposure, covariates, and outcomes among practice attenders and nonattenders. Robust effect implies insensitivity to bias introduced by varying degrees of health care contact | 1,825,694 | 125,472/10,460,654 | 1.16 (1.14-1.18) | 2,086,308 | 100,225/11,882,522 | 1.19 (1.16-1.21) |
| Sensitivity analysis 4 | Repeating the primary analysis on redefined cohorts with a less-restrictive atopic eczema definition: atopic eczema diagnosis was ascertained using only atopic eczema diagnostic codes, with no requirement for a therapeutic code | To explore the sensitivity of the results to the definition of atopic eczema (eg, those with childhood atopic eczema may have been erroneously excluded from the primary analysis if they switched practice in adulthood, and did not require further treatments) | 2,514,107 | 173,793/14,708,229 | 1.07 (1.05-1.09) | 2,838,141 | 135,719/16,515,260 | 1.10 (1.08-1.12) |
| Sensitivity analysis 5 | Repeating the primary analysis on redefined cohorts with a less-restrictive definition for those without atopic eczema (unexposed): individuals with an atopic eczema diagnosis but without 2 further eczema treatments were considered not to have atopic eczema, and could therefore be included in the pool of unexposed participants. The cohort of patients with atopic eczema remained the same (ie, eczema was defined as having at least 1 diagnostic code and 2 treatment codes) | To explore the sensitivity of the results to the definition of atopic eczema | 2,002,613 | 135,552/11,329,039 | 1.14 (1.12-1.16) | 2,267,537 | 107,660/12,771,273 | 1.16 (1.14-1.19) |
| Sensitivity analysis 6 | Additionally adjusting for ethnicity (white, South Asian, black, other, or mixed, identified from CPRD and HES data). Analysis was restricted to those registered in 2006 or later, because ethnicity recording before 2006 is selective, and of low qualityE62 | To examine whether the omission of ethnicity from the primary analysis may have introduced bias, because reliable ethnicity data exists only for that period | 276,853 | 8,251/649,041 | 1.16 (1.06-1.26) | 340,161 | 8,481/80,4170 | 1.29 (1.18-1.40) |
PYAR, Person-years at-risk.
All models were fitted to patients with complete data for all included variables. Sets without at least 1 exposed and 1 unexposed were excluded. HRs were estimated from a Cox regression model with current age as the underlying timescale, stratified by matched set (sex, age, and general practice), and adjusted for current calendar period (years: 1998-2001, 2002-2006, 2007-2011, and 2012-2016,) and quintiles of IMD at cohort entry.
∗
All HRs are for the outcome (ie, depression/anxiety) among those with atopic eczema, compared with those without atopic eczema.