Study population and design

We used data from The Maastricht Study, an observational prospective population-based cohort study. The rationale and methodology have been described previously.^{Reference Schram, Sep, van der Kallen, Dagnelie, Koster and Schaper15} In brief, the study focuses on the aetiology, pathophysiology, complications and comorbidities of type 2 diabetes mellitus (T2DM), heart disease and other chronic conditions, and is characterised by an extensive phenotyping approach. All individuals aged between 40 and 75 years and living in the southern part of the Netherlands were eligible for participation. Participants were recruited through mass media campaigns, the municipal registries and the regional Diabetes Patient Registry via mailings. Recruitment was stratified according to known T2DM status, with an oversampling of individuals with T2DM, for reasons of efficiency. Baseline data were collected between November 2010 and January 2018. Lag time between magnetic resonance imaging (MRI) and depression assessment at baseline was 102 days (s.d. = 120).

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human participants were approved by the institutional Medical Ethical Committee (NL31329.068.10) and the Minister of Health, Welfare and Sports of the Netherlands (Permit 131088-105234-PG). All participants gave written informed consent.

For the current analysis complete data was available from 4653 participants for cross-sectional analysis and 4154 participants for longitudinal analysis. Supplementary Figure 1 available at https://doi.org/10.1192/bjp.2023.143 shows the flow chart of the study population.

Brain MRI

Brain images were acquired on a 3T clinical magnetic resonance scanner (MAGNETOM Prismafit, Siemens Healthineers GmbH) located at a dedicated scanning facility (Scannexus, Maastricht, the Netherlands) using a head/neck coil with 64 elements for parallel imaging. The MRI protocol included a three-dimensional T ₁-weighted (T ₁w) magnetisation-prepared rapid acquisition gradient echo (MPRAGE) sequence (repetition time/inversion time/echo time (TR/TI/TE) 2300/900/2.98 ms, 176 slices, 256 × 240 matrix size, 1.0 mm cubic reconstructed voxel size); and a fluid-attenuated inversion recovery (FLAIR) sequence (TR/TI/TE 5000/1800/394 ms, 176 slices, 512 × 512 matrix size, 0.49 × 0.49 × 1.0 mm reconstructed voxel size).

Brain segmentation was performed with FreeSurfer v6.0^{Reference Fischl16} using both T ₁w and FLAIR images as input. The arguments ‘-FLAIRpial’ and ‘-3T’ were used to optimise segmentation quality. Brain segmentations with insufficient quality, i.e. Euler numbers below 1.5 quartile (−80 for left hemisphere and −68 for right hemisphere) were excluded.^{Reference Monereo-Sánchez, de Jong, Drenthen, Beran, Backes and Stehouwer17} Hippocampal subfields^{Reference Iglesias, Augustinack, Nguyen, Player, Player and Wright18} were segmented using multispectral segmentation, yielding hippocampus total volume and 12 hippocampal subfields per hemisphere (Supplementary Table 1). All extracted volumes were z-transformed prior to statistical analysis with respect to the distribution in the complete sample (n = 4643). Results are depicted using hippocampal subfields maps, a legend for these maps can be found in Supplementary Figure 2.

Depression

Depressive symptoms were assessed by a validated Dutch version of the 9-item Patient Health Questionnaire (PHQ-9)^{Reference Kroenke, Spitzer and Williams19} both at baseline and follow-up. Follow-up data was collected annually over a period of 7 years, i.e. each participant was asked to complete the PHQ-9 questionnaire once every year, up to 7 years. The PHQ-9^{Reference Kroenke, Spitzer and Williams19} is a self-administered questionnaire that assesses the presence of the nine symptoms for the DSM-IV criteria for a major depressive disorder on a four-point Likert scale ranging from 0 ‘not at all’ to 4 ‘nearly every day’.²⁰ When one or two items were missing, the total score was calculated as 9 × (total points/9 − number of missing items) and rounded to the nearest integer. When more items were missing, the total score was scored as missing. A cut-off score of ≥10 is most often used as a dichotomous scoring system for defining clinically relevant depressive symptoms, with a good sensitivity (88%) and specificity (78%).^{Reference Pettersson, Boström, Gustavsson and Ekselius21} The internal consistency of the PHQ-9 in The Maastricht Study was good (Cronbach's alpha = 0.82 without T2DM, and 0.87 with T2DM).^{Reference Janssen, Köhler, Stehouwer, Schaper, Dagnelie and Sep22}

There was a time lag between the baseline data collection and the date of the MRI scan. Therefore, the PHQ-9 score obtained closest to the date of the MRI scan, regardless of whether the assessment was before or after the scan, was chosen as the baseline score for each individual. Subsequent assessments were labelled as follow-up 1, follow-up 2, and so forth, based on the order in which they occurred after the baseline assessment.

Here, we use the term ‘prevalent depressive symptoms’ to indicate the use of PHQ-9 scores as a continuum at baseline. We use the term prevalent depression to indicate clinically relevant depressive symptoms (PHQ-9 ≥ 10) at baseline. We subdivided prevalent depression according to its course as:

1. (a) ‘prevalent depression with a chronic course’ i.e. clinically relevant depressive symptoms (PHQ-9 ≥ 10) at baseline and clinically relevant depressive symptoms (PHQ-9 ≥ 10) at at least one follow-up time; and

2. (b) ‘prevalent depression with a transient course’ i.e. clinically relevant depressive symptoms (PHQ-9 ≥ 10) at baseline and no clinically relevant depressive symptoms (PHQ-9 < 10) during follow-up.

We use the term ‘incident depression’ to indicate no clinically relevant depressive symptoms (PHQ-9 < 10) at baseline and presence of clinically relevant depressive symptoms (PHQ-9 ≥ 10) at at least one follow-up time. We subdivided incident depression according to its course as:

1. (a) ‘Incident depression with a chronic course’ i.e. no clinically relevant depressive symptoms (PHQ-9 < 10) at baseline and clinically relevant depressive symptoms (PHQ-9 ≥ 10) at two or more follow-up moments; or

2. (b) ‘incident depression with a transient course’, i.e. no clinically relevant depressive symptoms (PHQ-9 < 10) at baseline and clinically relevant depressive symptoms (PHQ-9 ≥ 10) at one follow-up.

We used the term, ‘no depression’ as the comparison group, and include those participants with no clinically relevant depressive symptoms (PHQ-9 < 10) at baseline and no clinically relevant depressive symptoms (PHQ-9 < 10) at follow-up.

General characteristics and covariates

General characteristics and covariates were measured at baseline. Educational level (low, intermediate, high), history of cardiovascular disease (CVD), smoking status (never, current, former), alcohol consumption (none, low, high) were assessed by questionnaires.^{Reference Schram, Sep, van der Kallen, Dagnelie, Koster and Schaper15} We measured, height, weight, waist circumference, office blood pressure, plasma lipid profile, and 24 h urinary albumin excretion (twice) as described elsewhere.^{Reference Schram, Sep, van der Kallen, Dagnelie, Koster and Schaper15}

To determine T2DM status, all participants (except those who used insulin) underwent a standardised seven-point oral glucose tolerance test after an overnight fast. Glucose metabolism status was defined according to the World Health Organization 2006 criteria.²³ Participants were considered to have T2DM if they had a fasting blood glucose ≥7.0 mmol/L or a 2 h post load blood glucose ≥11.1 mmol/L or used oral glucose-lowering medication or insulin. Cholesterol lowering medication, glucose-lowering medication and use of antidepressants was assessed in a medication interview at baseline where generic name, dose and frequency were registered.

Statistical analyses

General characteristics of the study population were evaluated using independent t-tests, or χ²-tests when appropriate.

A total of five research questions (RQs) were asked. Are hippocampal subfield volumes associated with:

1. (a) prevalent depressive symptoms (RQ1);

2. (b) clinically relevant depressive symptoms (RQ2);

3. (c) clinically relevant depressive symptoms according to its course (chronic and transient) (RQ3);

4. (d) incident clinically relevant depressive symptoms (RQ4); and

5. (e) incident clinically relevant depressive symptoms according to its course (chronic and transient) (RQ5)?

We used negative binomial regression on data from n = 4643 participants to answer RQ1; logistic regression on data from n = 4643 participants for RQ2; multimodal logistic regression on data from n = 4174 participants for RQ3; and Cox proportional hazards regression with time to event on the time axis on data from n = 4174 participants for RQ4 and RQ5. An overview of the research questions, participants and groups can be seen in Supplementary Figure 1.

Associations were adjusted for potential confounders in two different models: model 1, adjusted for total brain volume (when analysing total hippocampal volumes), or for total hippocampal volume (when analysing hippocampal subfields), MRI lag time, age and gender; and model 2, additionally adjusted for T2DM status, education level, waist circumference, history of CVD, total-to-high-density lipoprotein cholesterol ratio, use of alcohol and smoking status.

We studied the left and right hemispheres separately, and analysed 1 total hippocampal volume and 12 hippocampal subfields for each hemisphere. The hippocampal subfields are correlated with each other (Supplementary Table 2). For this reason, correction for multiple comparisons was done in accordance with the matrix spectral decomposition method.^{Reference Nyholt24} Based on the resulting eigenvalues, the obtained effective number was n = 13, therefore alpha was set at 0.05/13 = 0.0039.

Several sensitivity analyses were performed based on the fully adjusted model (model 2):

1. (a) we excluded individuals with type 2 diabetes to assess whether they drive the observed associations;

2. (b) we adjusted for antidepressant medication use;

3. (c) we excluded participants who used antidepressant medication;

4. (d) to restrict analyses to ‘de novo’ depression, we excluded participants who had a history of major depressive disorder diagnosis (assessed through the Mini-International Neuropsychiatric Interview^{Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs and Weiller25}) before baseline;

5. (e) we additionally adjusted for cognitive status using Mini-Mental State Examination score.^{Reference Folstein, Folstein and McHugh26}

Finally, we tested whether these associations differed according to gender, and T2DM status, by use of interaction analyses.

All statistical analyses were performed in R 4.0.2 (2020-06-22); analytic code is available on request from the corresponding author.