1 INTRODUCTION

Major depressive disorder (MDD) and generalized anxiety disorder (GAD) are the most common psychiatric disorders worldwide (Kessler et al., 2005; Lim et al., 2018). Around 11% of the world's population experience clinically relevant symptoms within their lifetime (Lim et al., 2018) and they co-occur simultaneously or sequentially in 50%–80% of cases (Brown et al., 2001; Jacobson & Newman, 2017; Kroenke et al., 2007; Moffitt et al., 2007). MDD and GAD are heterogeneous conditions, meaning affected individuals can present with unique symptom combinations (Fried & Nesse, 2015a, 2015b). In research, questionnaire-derived sum scores are commonly used to identify cases or measure disorder severity (Fried & Nesse, 2015b; Jokela et al., 2019; Ohannessian et al., 2017). Sum scores are a symptom count, which assumes all symptoms equally and consistently contribute to a disorder (Fried & Nesse, 2015b). They also lack information on which symptoms are present for an individual and how those symptoms interact (Jokela et al., 2019). For instance, four underlying factors of generalized anxiety and depressive symptoms were identified in the UK Biobank: anxiety symptoms, psychomotor-cognitive impairment, neurovegetative states, and mood symptoms (Jermy et al., 2020). Exploring patterns of symptom co-occurrence instead of using sum scores may identify diagnostic subtypes of MDD, GAD, or their comorbid presentation, helping to refine diagnoses (Eeden et al., 2019).

The presentation of MDD symptoms varies across ages (Schaakxs et al., 2017). Older people have been shown to more often report somatic symptoms (Hegeman et al., 2012; Miloyan & Pachana, 2016; Schaakxs et al., 2017; Zhao et al., 2018). However, the age-dependent presentation of symptoms is difficult to detect using questionnaire-derived sum scores. While depression sum scores remain consistent across adulthood, older people are more likely to report fatigue, psychomotor agitation, and sleep problems, whereas younger people are more likely to report irritability, concentration problems, and anxiety (Schaakxs et al., 2017). Similarly, endorsement of depression symptoms can vary by sex. Women are more likely to endorse somatic symptoms of fatigue, muscle tension, sleep problems, and appetite problems, whereas men are more likely to endorse suicidal ideation (Fried et al., 2014; Vesga-López et al., 2008). It is unclear whether symptom level age and sex variability extends to GAD symptomology. Given the use of brief measures of generalized anxiety and depression symptoms as the criteria for entry into the UK NHS Improving Access to Psychological Therapies (IAPT) service, it is important to understand how symptom presentations assessed by these measures might vary with age and sex.

1.1 Aims

In the present study, we aim to (a) identify latent factors composed of MDD and GAD symptoms in participants from the Genetic Links to Anxiety and Depression (GLAD) Study and (b) assess how individual MDD and GAD symptoms and our identified factors were associated with age and sex.

2 METHODS 2.1 Study design

The GLAD Study is an ongoing study of anxiety and depression (Davies et al., 2019). Participants are recruited via an online platform from the general population and National Health Service organizations. Individuals above 16 years of age, living in the United Kingdom who have experienced depression or any anxiety disorder including GAD, social anxiety disorder, panic disorder, agoraphobia, social phobia, and specific phobia are eligible. Participants complete the online questionnaire and donate a saliva sample. All participants provided full consent to take part and to the long-term storage of their data. Ethical approval was obtained from the London-Fulham Research Ethics Committee (REC reference: 18/LO/1218).

2.2 Participants

Our analyses included 35,637 individuals who completed the online questionnaire before May 19, 2020 and had no missing data on age, sex, Generalized Anxiety Disorder (GAD-7), or Patient Health Questionnaire (PHQ-9). Participants reported their biological sex, ethnicity, highest education level, and mental health diagnoses. Age divided by 10 was used in analyses to aid odds ratio (OR) interpretation. Participants’ age ranged from 16 to 93 years, with a mean of 38.1 years (SD = 14.4). The majority of the sample was female (79.6%), White (94.3%), and highly educated (Table 1). Overall, 88.2% of participants self-reported an MDD diagnosis and 76.6% a GAD diagnosis.

Table 1. Demographic characteristics of the Genetics Links to Anxiety and Depression (GLAD) sample (N = 35,637) Male (N = 7265) Female (N = 28372) Total (N = 35637) N % N % N % Age Mean (SD) 43 (14.58) 37 (14.16) 38 (14.41) Min 16 16 16 Max 93 93 93 Age-group 16–25 years 1031 2.89 7210 20.23 8241 23.12 26–35 years 1582 4.44 7749 21.74 9331 26.18 36–45 years 1515 4.25 5162 14.48 6677 18.73 46–55 years 1633 4.58 4734 13.28 6367 17.86 56–65 years 1044 2.93 2637 7.40 3681 10.33 66–75 years 403 1.13 800 2.24 1203 3.37 76–85 years 52 0.15 74 0.21 126 0.36 86–95 years 5 0.01 6 0.02 11 0.03 Ethnicity White 6856 19.24 26781 75.15 33637 94.39 Mixed 138 0.39 743 2.08 881 2.47 Asian or Asian British 114 0.32 359 1.01 473 1.33 Black or Black British 32 0.09 143 0.40 175 0.49 Arab 8 0.02 25 0.07 33 0.09 Other 91 0.26 254 0.71 345 0.97 Highest education level GCSE/CSE 919 2.58 3385 9.50 4304 12.08 NVQ 690 1.94 2293 6.43 2983 8.37 A-levels 1463 4.11 6520 18.30 7983 22.41 University 3876 10.88 15114 42.41 18990 53.29 Self-reported diagnosis MDDa 6374 17.89 25037 70.26 31411 88.15 MDD onlyb 800 2.24 2915 8.18 3715 10.42 GADa 5158 14.47 22143 62.13 27301 76.60 GAD onlyb 211 0.59 1030 2.89 1241 3.48 Anxiety disordera 2121 5.95 8658 24.29 10779 30.24 Anxiety disorder onlyb 2027 5.69 8187 22.97 10214 28.66 Abbreviations: CSE, certificate of secondary education; GAD, generalized anxiety disorder; GCSE, general certificate of secondary education; MDD, major depressive disorder; NVQ, national vocational qualification. aIndividuals who reported the diagnosis regardless of comorbidities. bIndividuals who only reported that diagnosis with no comorbidities. 2.3 Measures 2.3.1 Depression and generalized anxiety symptoms

We measured current MDD symptoms with the PHQ-9 (nine items; Kroenke et al., 2001, 2010) and current GAD symptoms with the GAD-7 (seven items; Spitzer et al., 2006). These are commonly used self-report scales based on the Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) diagnostic criteria (American Psychiatric Association, 2013). For both scales, individuals rated the frequency of symptoms experienced over the past 2 weeks. Items were rated on a four-point Likert-type scale with scores ranging from 0 (not at all) to 3 (nearly every day). All item scores were summed to create a severity sum score, with values ranging from 0 to 27 for the PHQ-9 and 0 to 21 for the GAD-7. Both scales were psychometrically valid with high test–retest reliability (GAD: r = .83; PHQ: r = .84), and internal consistency (GAD: alpha = .89; PHQ: alpha = .83; Löwe et al., 2008; Spitzer et al., 2006; Thorp et al., 2019).

2.4 Statistical analyses 2.4.1 Exploratory and confirmatory factor analyses

Polychoric correlation matrices were computed for all ordinal items (Holgado–Tello et al., 2008). To check for singularity/multicollinearity of items, the matrix was examined for values <0.30 and >0.90. The matrix determinant, Bartlett's Test of Sphericity (Bartlett, 1950), Ordinal alpha (Gadermann et al., 2012), and the Kaiser–Meyer–Olkin (KMO) statistic (Kaiser, 1974) were computed to assess whether the data were fit for exploratory factor analysis (EFA). Parallel analysis (Horn, 1965), Very Simple Structure (VSS; Revelle & Rocklin, 1979), and Velicer's Minimum Average Partial (MAP) criterion (Velicer, 1976) estimated the preliminary number of factors. See Supporting Information Material A and B.

'EFA was performed in 70% of the sample using the weighted least squares method in the “psych” R package (Revelle, 2017), which is preferred for ordinal data (Forero et al., 2009; Lee et al., 2012). Factors were allowed to correlate using oblimin rotation. The following fit criteria indicated good fit: root mean square error of approximation (RMSEA) ≤ 0.05, Tucker–Lewis Index (TLI) ≥ 0.95, standardized root mean square residuals (SRMR) ≤ 0.05, and a smaller Bayesian Information Criterion (BIC) relative to other models (Hu & Bentler, 1999). Items were retained in a factor for factor loadings of >0.3 and greater than loadings on all other factors. Where multiple models showed adequate fit, the model with factors that encompass the greatest number of items was chosen. To validate the EFA-derived model, confirmatory factor analysis (CFA) was conducted in the remaining 30% of the sample using the “lavaan” R package (Rosseel, 2012). Standardized fit statistics were interpreted (Hu & Bentler, 1999; Schreiber et al., 2006) and the comparative fit index (CFI) ≥ 0.95 was considered good fit. The CFA was computed in the full sample to provide overview fit statistics.

Two sensitivity analyses were conducted. First, items with a low loading (~0.3) on all factors were sequentially removed. If item removal did not substantially improve model fit, the item was retained. Second, the final model was computed for males and females separately to assess sex differences in factor structure.

2.4.2 Age and sex-related variability generalized anxiety and depression

We conducted three sets of regression analyses with age and sex. First, we fitted linear regressions associating age/10 with PHQ-9 and GAD-7 sum scores while controlling for sex. Each item was transformed into a binary variable to indicate the presence or absence of the symptom; 0 was coded as no symptom, and 1–3 were coded as symptom present. Second, we fitted logistic regressions associating dichotomized symptoms with age per 10 years and sex. Sum scores were also controlled for, to assess only the occurrence of the individual symptom and account for cumulative symptom disorder severity (Schaakxs et al., 2017). For age, an OR > 1 indicated an association with being 10 years older and OR < 1 indicated an association with being 10 years younger. For sex, OR > 1 indicated an association with being female while an OR < 1 indicated an association with being male. Third, we regressed the CFA-derived factor scores on age and sex controlling for disorder severity, to identify symptom groups that may vary across age and between the sexes.

To investigate nonlinear relationships between age and generalized anxiety/depression symptoms, logistic regression models were computed using categorical age. Age was categorized into year groups of 16–23, 24–31, 32–39, 48–55, 56–63, and >64 to avoid a reduction in power. The middle group of 40–47 years was used as the reference group. Two post hoc sensitivity analyses were conducted. First, individuals who participated during the COVID-19 pandemic were excluded at three intervals: January 31, 2020 (first UK case; N = 2456), March 1, 2020 (higher awareness; N = 1222), and March 23, 2020 (first UK lockdown; N = 342). Second, regression analyses were replicated accounting for highest education level as a proxy for socioeconomic status in the model.

The false discovery rate (FDR) multiple testing correction was applied to all symptom and factor level analyses separately (Benjamini & Hochberg, 1995; see Supporting Information Material C). All analyses were conducted using R version 4.0.2. All R code can be found in https://github.com/knthompson26/Age-sex-GAD7-PHQ9-GLAD.

3 RESULTS 3.1 EFA

The item “Worrying too much about different things” was excluded from EFA due to a polychoric correlation of 0.92 with the item “Difficulty controlling worrying.” For information on item endorsement and correlation structure, see Supporting Information Material D. EFA in 70% of the sample (N = 24,946) showed a four-factor solution best fit the data (Table 2). The RMSEA was slightly above the recommended threshold (Hu & Bentler, 1999). All other fit statistics indicated good model fit. All factors included two to five items, factor correlations ranged from 0.42 to 0.72 and the four-factor model explained 70% of the variance in the data. We labeled the four factors according to the loaded items: mood, worry, somatic, and motor symptoms (Figure 1). Factors did not neatly split into GAD and MDD symptoms, for example, motor symptoms contained one symptom from both the PHQ-9 and GAD-7. Five and six-factor models showed slightly better fit; however, these solutions were unsuitable as they included factors with zero items given the loading cut-off of 0.3. See Table S1 for EFA results with all items.

Exploratory factor analysis of GAD-7 and PHQ-9 items in the Genetic Links to Anxiety and Depression (GLAD) Study (N = 35637; 15 items). (a) The path diagram, item factor loadings, and between-factor correlations for the four factors of worry symptoms, mood symptoms, somatic symptoms, and motor symptoms are shown. (b) The loading strength for every item on each of the identified factors are shown. Dark blue indicates a positive factor loading, white indicates no factor loading, and red indicates a negative factor loading. GAD-7, Generalized Anxiety Disorder; PHQ-9, Patient Health Questionnaire

Table 2. Model fit statistics for exploratory factor analysis of one to six factors in the Genetic Links to Anxiety and Depression (GLAD) sample (N = 35,637; 15 items) Number of factors df RMSEA (≤ 0.06) RMSEA 90% CI TLI (≥ 0.95) BIC SRMR (≤ 0.08) Cumulative variance Minimum item loading 1 90 0.175 [0.174, 0.176] 0.736 67,991 0.09 0.56 15 2 76 0.126 [0.125, 0.127] 0.863 29,469 0.05 0.63 6 3 63 0.104 [0.102, 0.105] 0.907 16,359 0.04 0.67 0 4 51 0.076 [0.075, 0.078] 0.950 6907 0.02 0.70 2 5 40 0.055 [0.053, 0.057] 0.974 2665 0.02 0.72 0 6 30 0.040 [0.038, 0.042] 0.986 931 0.01 0.74 0 Note: The cut off for each statistic to signify “good” fit is listed in each header (Hu & Bentler, 1999). The model with the lowest BIC is preferred. Cumulative variance is given as an indicator of the variance explained between items by the number of factors in each model. The best fitting and chosen model is indicated in bold. Abbreviations: BIC, Bayesian information criterion; df, degrees of freedom; RMSEA, root mean square error of approximation; SRMR, standardized root mean square residuals; TLI, Tucker–Lewis fit index. 3.1.1 EFA sensitivity analyses

Concentration problems and subsequently irritability (items with the lowest loadings) were dropped from the EFA in an attempt to improve model fit. This solution then showed only a very slight improvement, thus the full model was retained (Table S2). This model was then fitted separately in males and females to check for sex differences in factor loadings. In males, the concentration problems item loaded onto the somatic symptoms factor rather than the mood factor as seen in the full and female only model (Figure S5).

3.2 CFA

The CFA model is predefined to that identified by EFA, which provides a more stringent test of model fit compared with EFA (Thompson, 2004). The CFA in the remaining 30% of the sample (N = 10,691) confirmed that the four-factor model fit the data well. The RMSEA (0.068, 95% confidence interval [CI]: 0.067, 0.091) was still slightly above the threshold for good fit (Hu & Bentler, 1999), although lower than that for the EFA. The TLI (0.993), CFI (0.994), and SRMR (0.046) all indicated good fit (Hu & Bentler, 1999). The CFA was then rerun in the full sample to provide an overall model (CFI = 0.99, TLI = 0.99, RMSEA = 0.07, SRMR = 0.04). Factor scores for each individual were computed from this final model and used in the latent factor regression analyses.

3.3 Age and sex-related variation in generalized anxiety and depression 3.3.1 Sum scores

Women reported higher depression and generalized anxiety sum scores compared with men. When controlling for sex, age was significantly associated with a 0.95 lower PHQ-9 (p < .001) and a 0.98 lower GAD-7 sum score (p < .001).

3.3.2 Individual symptoms

Eight symptoms were associated with younger age and four with older age, when controlling for sex, generalized anxiety severity and depression severity (Figure 2a). Irritability and suicide ideation were most strongly associated with younger age, followed by restlessness, weight/appetite problems, motor problems, little energy, feeling anxious, and concentration problems. Sleep change was most strongly associated with older age, followed by difficulty controlling worrying, worrying too much and trouble relaxing. See Table S3 for full OR and CI values.

Age- and sex-related variation in generalized anxiety and depression in the Genetic Links to Anxiety and Depression (GLAD study; N = 35,637). (a) Odds ratios (OR) and 95% confidence intervals (CI) for the association between all GAD-7 and PHQ-9 items and age (per 10 years) in blue and sex in yellow are shown. (b) Standardized estimates and 95% confidence intervals for the association between four factors of GAD-7 and PHQ-9 symptoms and age (in blue) and sex (in yellow) are shown. For age, points on the right of each panel indicate an association with older age per 10 years and points to the left indicate an association with younger age per 10 years. For sex, points to the right of each panel indicate an association with being female, to the left indicate an association with being male. Filled circle points indicate significant associations while accounting for multiple testing using FDR correction. Transparent circle points indicate nonsignificant associations while accounting for multiple testing using FDR correction. GAD-7, Generalized Anxiety Disorder; FDR, false discovery rate; PHQ-9, Patient Health Questionnaire

More symptoms were associated with being male than female (Figure 2a). Weight/appetite problems, little energy, sleep change, difficulty controlling worrying, and worrying too much were more likely endorsed by females, while suicide ideation, anhedonia, depressed mood, restlessness, worthlessness, concentration problems, and motor problems were more likely endorsed by males. See Table S3 for full OR and CI values.

These associations were replicated using categorical age groups. Categorical age measurement also identified symptom occurrence that cannot be captured by continuous age. Individuals who were 16–23 or over 64 years were less likely to report concentration problems, little energy, trouble relaxing, weight or appetite problems and worthlessness (Figure S6). This suggests a nonlinear relationship, where individuals of middle age are more likely to report these symptoms.

3.3.3 Latent factors

Table 3 displays age and sex associations with each symptom categorized by the four corresponding factors. When controlling for sex, generalized anxiety severity and depression severity, younger age per 10 years was significantly associated with 0.008 lower motor symptoms (p < .001), and with 0.003 lower somatic symptoms (p < .001; Figure 2b). When controlling for age, generalized anxiety severity and depression severity, males were more likely to report mood (β = −.04, p < .001), and motor symptoms (β = −.02, p < .001), while females were more likely to report somatic symptoms (β = .06, p < .001; Figure 2b).

Table 3. Age and sex-related variability in the four factors of Generalized Anxiety Disorder 7-item (GAD-7) and Patient Health Questionnaire-9 (PHQ-9) symptoms in the Genetic Links to Anxiety and Depression (GLAD) Study (N = 35,637) Factor Age variability Sex variability Symptom Age variability Sex variability Mood symptoms -