Socioeconomic Indicators of Treatment Prognosis for Adults With Depression

This systematic review and meta-analysis investigates the socioeconomic factors associated with depression treatment outcome regardless of treatment type.

If applicable, describe how any studies for which IPD were not available were dealt with. This should include whether, how and what aggregate data were sought or extracted from study reports and publications (such as extracting data independently in duplicate) and any processes for obtaining and confirming these data with investigators.

Data items 11
Describe how the information and variables to be collected were chosen. List and define all study level and participant level data that were sought, including baseline and follow-up information. If applicable, describe methods of standardising or translating variables within the IPD datasets to ensure common scales or measurements across studies. 6,7, S Tab 3

IPD integrity A1
Describe what aspects of IPD were subject to data checking (such as sequence generation, data consistency and completeness, baseline imbalance) and how this was done.

3, 8, S Tab 3
Risk of bias assessment in individual studies. 12 Describe methods used to assess risk of bias in the individual studies and whether this was applied separately for each outcome. If applicable, describe how findings of IPD checking were used to inform the assessment. Report if and how risk of bias assessment was used in any data synthesis. 3,9 Specification of outcomes and effect measures 13 State all treatment comparisons of interests. State all outcomes addressed and define them in detail. State whether they were pre-specified for the review and, if applicable, whether they were primary/main or secondary/additional outcomes. Give the principal measures of effect (such as risk ratio, hazard ratio, difference in means) used for each outcome. Synthesis methods 14 Describe the meta-analysis methods used to synthesise IPD. Specify any statistical methods and models used. Issues should include (but are not restricted to):  Use of a one-stage or two-stage approach.  How effect estimates were generated separately within each study and combined across studies (where applicable).  Specification of one-stage models (where applicable) including how clustering of patients within studies was accounted for.  Use of fixed or random effects models and any other model assumptions, such as proportional hazards.  How (summary) survival curves were generated (where applicable).  Methods for quantifying statistical heterogeneity (such as I 2 and  2 ).  How studies providing IPD and not providing IPD were analysed together (where applicable).  How missing data within the IPD were dealt with (where applicable).

Exploration of variation in effects A2
If applicable, describe any methods used to explore variation in effects by study or participant level characteristics (such as estimation of interactions between effect and covariates). State all participant-level characteristics that were analysed as potential effect modifiers, and whether these were pre-specified. [6][7][8] Risk of bias across studies 15 Specify any assessment of risk of bias relating to the accumulated body of evidence, including any pertaining to not obtaining IPD for particular studies, outcomes or other variables. 9 Additional analyses 16 Describe methods of any additional analyses, including sensitivity analyses. State which of these were pre-specified. 9

Results
Study selection and IPD obtained 17 Give numbers of studies screened, assessed for eligibility, and included in the systematic review with reasons for exclusions at each stage. Indicate the number of studies and participants for which IPD were sought and for which IPD were obtained. For those studies where IPD were not available, give the numbers of studies and participants for which aggregate data were available. Report reasons for non-availability of IPD. Include a flow diagram. Study characteristics 18 For each study, present information on key study and participant characteristics (such as description of interventions, numbers of participants, demographic data, unavailability of outcomes, funding source, and if applicable duration of follow-up). Provide (main) citations for each study. Where applicable, also report similar study characteristics for any studies not providing IPD. Table 1 IPD integrity A3 Report any important issues identified in checking IPD or state that there were none.

3, S Tab 3
Risk of bias within studies 19 Present data on risk of bias assessments. If applicable, describe whether data checking led to the up-weighting or downweighting of these assessments. Consider how any potential bias impacts on the robustness of meta-analysis conclusions. Results of individual studies 20 For each comparison and for each main outcome (benefit or harm), for each individual study report the number of eligible participants for which data were obtained and show simple summary data for each intervention group (including, where applicable, the number of events), effect estimates and confidence intervals. These may be tabulated or included on a forest plot.

Tab 3-4, 8-19
Results of syntheses 21 Present summary effects for each meta-analysis undertaken, including confidence intervals and measures of statistical heterogeneity. State whether the analysis was pre-specified, and report the numbers of studies and participants and, where applicable, the number of events on which it is based.
Provide a description of the direction and size of effect in terms meaningful to those who would put findings into practice.

Risk of bias across studies 22
Present results of any assessment of risk of bias relating to the accumulated body of evidence, including any pertaining to the 9, S Tab availability and representativeness of available studies, outcomes or other variables.

Additional analyses 23
Give results of any additional analyses (e.g. sensitivity analyses). If applicable, this should also include any analyses that incorporate aggregate data for studies that do not have IPD. If applicable, summarise the main meta-analysis results following the inclusion or exclusion of studies for which IPD were not available.

Summary of evidence 24
Summarise the main findings, including the strength of evidence for each main outcome.

2,3,10
Strengths and limitations 25 Discuss any important strengths and limitations of the evidence including the benefits of access to IPD and any limitations arising from IPD that were not available.

Conclusions 26
Provide a general interpretation of the findings in the context of other evidence.

12-13
Implications A4 Consider relevance to key groups (such as policy makers, service providers and service users). Consider implications for future research.

Funding 27
Describe sources of funding and other support (such as supply of IPD), and the role in the systematic review of those providing such support.
Additional Information on the Prognostic indicators 1) Employment Status (available in 9 studies, n=4864) -An eight category variable capturing this existed in the dataset: i) employed full-time; ii) employed part-time; iii) houseperson; iv) retired; v) student; vi) unemployed job seeker; vii) unemployed due to ill health; viii) other. However, based on previous work with the data which included the studies included here, there were insufficient numbers of participants in some categories in some studies to use all eight categories above. For example, there was only one houseperson in the REEACT study, five in the MIR study and nine in the COBALT study. There were also only four retirees in the IPCRESS study, and seven in the TREAD study. So, these categories were collapsed together to form a three category variable: i) employed (including full time and part time employed); ii) unemployed (including unemployed job seeker and unemployed due to ill health); and iii) not seeking employment (including all other categories). 2) Financial Strain (available in seven studies, n=3656) -A five category variable capturing this existed in the data provided: i) Living comfortably; ii) Doing alright financially; iii) Just about getting by; iv) finding it difficult to make ends meet; v) very difficult to make ends meet. Again, there were low numbers in some categories within some of the studies, so this variable was collapsed into three categories: i) Doing OK Financially (living comfortably or doing alright financially); ii) just about getting by; iii) struggling financially (difficult or very difficult to make ends meet). 3) Housing Status (available in eight studies, n=4397) -a six category variable capturing housing tenancy status existed in the data: i) Homeowner, ii) Tenant, iii) Living with family/friends; iv) Hostel, v) Homeless, vi) Other. There were very few participants that reported being homeless or living in hostels in most of the studies, so this variable will be re-categorised into three categories: i) Homeowner; ii) Tenant; and iii) Other. 4) Highest Level of Educational Attainment (available in eight studies, n=3689) -a six category variable capturing this existed: i) Degree of above; ii) Foundation degree, higher national diploma, or equivalent; iii) A-level; iv) GCSE; v) Other qualifications (below the level of GCSEs); vi) No formal qualifications. There were no participants with "Other qualifications" in COBALT, and none with "No formal qualifications" in REEACT and MIR. This variable was therefore re-categorised into the following four categories: i) Bachelor's Degree or higher; ii) A-levels or Diplomas including Foundation Degrees; iii) GCSE; iv) Other qualifications (below the level of GCSEs) or no formal qualifications.
Additional Information on Calculation of 'Percentage Difference' primary outcome: To calculate the 'percentage difference' primary outcome, a new variable was created by calculating the natural logarithm of the depressive symptom measure scores for each participant at 3-4 months post-baseline, irrespective of the depressive measure used (i.e. the variable contained the natural logarithm of scores for participants that had completed the PHQ-9 and those that had completed the BDI-II). This variable was fitted as a continuous outcome variable in the regression models, and the exponent (e x ) of the coefficient for the prognostic variable (the socioeconomic variable) was then calculated to give the effect estimate per one-unit increase in that variable for the ordinal socioeconomic variables or comparing the stated category to the reference category (e.g. employed compared to unemployed). 95% confidence intervals were created with the exponentiated values. The percentage difference for the socioeconomic variable in each model was calculated as: 100*(the exponentiated coefficient -1). These percentage differences could then be compared with estimates for minimal clinically important differences on the PHQ-9 and BDI-II from previous research with primary care patients. 1

Data Extraction
Raw data were extracted for each study participant of all studies meeting inclusion criteria that agreed to provide individual patient data. Data were cleaned one study at a time, independently by two reviewers (JB and RS).

Data Integrity Checks
Integrity of all baseline and endpoint data for each study were checked with the study team and against details published about each study. The numbers of participants included in this dataset for four studies was slightly different from those in the published articles about the individual studies. This is because a very small number of cases were removed as they had missing data on over 75% of the variables at baseline, this resulted in two patients being removed from the IPCRESS study and one from the PANDA study. For the CADET study 54 participants withdrew after the study was completed so their data were not made available, and for the ITAS study there were complete data for 36 more participants than reported in the publications about that study.

Missing Data
Missing data were imputed using multiple imputation with chained equations (MICE) in Stata 16.0. Data not reasonably able to be log transformed to meet normality assumptions, were imputed using predictive mean matching (PMM) via a k-nearest neighbours approach as it is considered to be more appropriate for non-normal continuous variables 2 , here we used k=10. Linear regression was used for approximately normally distributed continuous variables, logistic regression models for binary variables, and ordinal and multinomial regression models for ordered and unordered categorical variables respectively. All imputation models were built using data on baseline and outcome variables following conventions 3 . Only variables with less than 50% missing data were imputed. All imputation models were run to produce 50 imputed datasets.

Patient and Public Involvement
Service user advisory groups of two primary care mental health services in central London and the expert service user researchers of the Service User Research Forum (SURF) were consulted for advice on the design, conduct, and dissemination of this study.

Details of preliminary searches, and additions, deviations and changes to protocols
We registered the process of finding studies and the research questions for this study on PROSPERO (CRD42019129512) and produced a protocol paper which was amended twice. Below we explain the amendments made and the process of finding studies and forming the dataset for this study.
We started this project by running some preliminary searches informed by consultation with a librarian at University College London, to identify studies of depression in primary care using the MEDLINE database via OVID, hand-searching through reference lists of existing systematic reviews and contacting a number of experts to enquire about unpublished or ongoing studies. No limits or filters were applied to the searches and no automatic updates were applied, although searches were re-run as detailed below. In these searches we found that the Clinical Interview Schedule, Revised version (CIS-R) 8 was the most commonly used comprehensive measures of depressive and anxiety symptoms and disorders in RCTs of depression set in primary care, among studies returned in the searches. Ten studies used the CIS-R at baseline to determine diagnosis (seven published and three protocols for studies that would soon be completed). Only one of the returned studies used the Schedules for Clinical Assessment in Neuropsychiatry (SCAN) 9 but did not meet all other inclusion criteria as no details were given on any socioeconomic factors and their associations with prognostic outcomes. 10 In addition, one study used the full Structured Clinical Interview for DSM (SCID) 11 but it too did not provide information on any socioeconomic factors measured at baseline, 12 see Supplementary Tables 6 and 7 for details. We therefore refined our preliminary searches to look for studies that used the CIS-R, and the use of CIS-R at baseline was made an inclusion criterion to minimize biases when harmonising data, 13 and ensure included studies had data on the depressive 'disorder characteristics' found to be independently associated with prognosis, 14 to meet the aim of ascertaining whether socioeconomic factors can add to prognostic information, in addition to routinely assessed clinical factors.
One of the senior investigators involved in this project (GL) was a lead or co-investigator on a number of trials that used the CIS-R and we made contact with the chief investigators of those studies to ask for in-principle agreement to access IPD from their trials. We then applied for funding for this project. Once funding was in place we registered our project on PROSPERO, at that point we had run two rounds of searches (preliminary searches and one set to inform our funding application), and we had obtained IPD data from four studies. We refined our searches by including other bibliographic databases and contacting other experts for missed studies, this helped us find further studies. We invited the chief investigators from each of those studies to join the project. We began to collect some further IPD from the studies that had agreed to take part. We then wrote up a protocol paper with information of what we would do with those IPD data once the dataset was complete. We ran further searches and found one more study just before initially submitting the protocol paper. The Protocol paper (including the searches) was peer-reviewed and we amended it post-review to give more details about this process. The protocol was then accepted for publication. It was amended once more when we decided that our choice of an I 2 threshold for considering problematic heterogeneity was too high, we dropped it from 80% to 75% in line with recommendations from Cochrane. We ran the final searches for studies meeting our inclusion criteria a few weeks before submitting this manuscript for publication and found no new studies meeting our criteria.
Our protocol paper provides information about all data we sought to extract from the included studies and all outcomes of interest. For the present study we were particularly interested in socioeconomic factors and potential confounders of the association between these factors and prognosis. We put together some exploratory directed acyclic graphs to consider what those confounders might be, and limited the data used for this study to those factors. Future studies using these data will consider the prognostic associations between other factors at baseline and prognosis. Further, for this study we amended our inclusion criteria slightly to exclude studies that did not include any measurement of socioeconomic factors at baseline, or that sampled only those in one category of any of those variables (e.g. all unemployed participants). There were two changes to the statistical analysis plan that should be noted: we did not include attrition as an outcome for the present study, and we also did not include conversions of scores on depressive symptom scales to the PROMIS T-score, 15 this was because here we included one study with a scale that could not be converted to the PROMIS (the GHQ-12).

Measure Details Scores and Cut-offs for Remission
The CIS-R 8 Consists of 14 symptom subsections scored 0-4 covering core features of depression, depressive thoughts (scored 0-5), fatigue, concentration/forgetfulness, and sleep, generalized anxiety, worry, irritability, obsessions, compulsions, health anxiety, somatic concerns, phobic anxiety (split into agoraphobia, social phobia, and specific phobia), and panic. A final section measures general health, impairment and weight change.
The total score ranges from 0-57 with a cut-off of ≥12 used to indicate likely common mental disorder, primary and secondary diagnoses using ICD-10 criteria are given as are binary indictors of diagnosis for all the disorders assessed. The duration of each type of problem is also assessed for the present episode (or subsyndromal episode) up to the point of completing the CIS-R. Duration items are measured in five categories: 1) less than two weeks; 2) between two weeks and six months; 3) between six months and one year; 4) between one and two years; and 5) more than two years.
Beck Depression Inventory 2 nd Edition (BDI-II) 16 Consists of 21 items to assess depressive symptoms, each item is scored 0-3.
There is a maximum score obtainable of 63, and a cut-off of ≥10 is used indicate significant symptoms of depression, scores of <10 are therefore used to indicate remission in those that were previously depressed/scored ≥10.
Patient Health Questionnaire 9-item version (PHQ-9) 17 This is a depression screening measure, with respondents asked to rate how often they have been bothered by each of the nine symptom items over the preceding two weeks. Each item is scored 0-3 There is a maximum score of 27 with a cut-off of ≥10 is used to indicate "caseness" for depression, a score of 9 or below for those that were previously depressed is therefore considered to indicate remission General Health Questionnaire (12-item version) (GHQ-12) 18 Consists of 12 items related to present and recent health over the "few weeks" prior to completion. Each item is related to depression or generalised anxiety, they are scored 0-0-1-1 for the four response options.
A cut-off of ≥2 is used to indicate the likely presence of common mental disorder, and so scores of <2 for those formally scoring above this would be considered to indicate remission Social Support Scaleadapted by authors of RCTs An 8-item instrument (the first seven of which are from the Health and Lifestyles Survey) assessing the degree to which participants rated the N/A 19 included in this IPD by adding one item to the Health and Lifestyles Survey Social Support Measure 20 .
social support of their friends and family in each of the following domains: 1) being accepted for who one is; 2) feeling cared about; 3) feeling loved; 4) feeling important to them; 5) being able to rely on them; 6) feeling well supported and encouraged by them; 7) being made to feel happy by them; and 8) feeling able to talk to them whenever one might like. Items are scored 1-3, with total scores ranging from 8-24; higher scores indicate higher levels of perceived social support. The authors of the Health and Lifestyles Survey suggested the maximum score for social support (which was 21 on that scale) indicated 'no lack of social support', scores between 18-20 indicated a 'moderate lack of social support', and scores of 17 or below indicated a 'severe lack of social support'. Life events: adapted by the authors of the Adult Psychiatric Morbidity Surveys 21 based on the Social Readjustment Rating Scale 22 Participants are asked to respond yes/no to whether they have suffered any of eight events within the last six months e.g. a death/bereavement; being physically attacked/injured; or going through a divorce/separation. Each item is scored yes (1) or no (0) and the total score is the sum of all the items.  7. ("primary care" or "general practice" or "general practitioner pr GP").af. 12. ("Clinical Interview Schedule" or "CIS-R" or "CISR" or "Revised clinical interview schedule" or "clinical interview schedule revised"