Perceived stress and diet quality in women of reproductive age: a systematic review and meta-analysis.

This was a systematic review and meta-analysis — not an experiment. The researchers asked: "Is there an association between how much psychological stress a woman perceives and the quality of her diet?" They did not test any intervention. Instead, they pooled data from existing observational studies that measured both stress (using questionnaires) and diet quality (using validated diet indices) in the same women at the same time point (cross-sectional) or across two time points (longitudinal).

The exposure was perceived psychological stress — how stressed a woman *feels*, not objective stressors like cortisol levels or life events. The outcome was diet quality, measured using pre-defined scoring systems like the Healthy Eating Index (HEI), Alternate Healthy Eating Index (AHEI), Dietary Approaches to Stop Hypertension (DASH) score, Dietary Quality Index for Pregnancy (DQI-P), and Dietary Guideline Adherence Index (DGAI). These indices assign points for eating recommended foods (fruits, vegetables, whole grains, lean protein) and penalise eating less healthy foods (saturated fat, sodium, added sugar).

The meta-analysis specifically looked at the correlation coefficient (r) between stress scores and diet quality scores — a number that ranges from −1 (perfect negative relationship: more stress = worse diet) through 0 (no relationship) to +1 (perfect positive relationship: more stress = better diet).

The meta-analysis included **5 studies with a total of 3,471 women** after sensitivity analysis removed two studies that were statistical outliers. The full systematic review included 24 studies (8 on diet quality, 16 on food intake/frequency) with a total of 3,982 participants across multiple countries: USA, UK, Australia, Canada, Japan, Iran, and several European nations.

All participants were:

Women aged 18–49 years (reproductive age range)

Healthy (no diagnosed depression, eating disorders, diabetes, heart disease, cancer, or other conditions that could confound the stress-diet relationship)

Non-pregnant at the time of most studies (though some included pregnant women with diet quality indices designed for pregnancy)

Recruited from community settings, university campuses, and healthcare clinics

The 16 food-intake studies (not included in the meta-analysis) covered a broader range of approximately 4,000–5,000 additional women, but exact totals are not reported because the studies were too heterogeneous to combine statistically.

**Stress measurement:** All studies used self-report questionnaires. The most common was the **Cohen Perceived Stress Scale (PSS)** — a 10-item or 14-item questionnaire asking how often in the past month you felt you could not control important things, felt nervous or stressed, or felt things were going your way. Scores typically range from 0–40 (10-item) or 0–56 (14-item), with higher scores = more perceived stress. No study measured biological stress markers like cortisol in blood, saliva, or hair.

**Diet quality measurement:** Eight different diet quality indices were used across the studies:

**Healthy Eating Index (HEI)** — scores 0–100, higher = better alignment with US Dietary Guidelines

**Alternate Healthy Eating Index (AHEI)** — scores 0–110, emphasises nuts, legumes, and moderate alcohol

**DASH Diet Index** — scores 0–90 or 0–11, based on the Dietary Approaches to Stop Hypertension pattern

**Dietary Quality Index for Pregnancy (DQI-P)** — pregnancy-specific, scores 0–100

**Dietary Guideline Adherence Index (DGAI)** — scores 0–100, based on adherence to dietary guidelines

For the 16 food-intake studies, diet was measured using food frequency questionnaires (FFQs), 24-hour dietary recalls, or food diaries, reporting intake of specific food groups (fruits, vegetables, sweets, fats, fast food) rather than an overall quality score.

**Study design:** This is a systematic review and meta-analysis of observational studies. The meta-analysis used a **random-effects model** to pool correlation coefficients (Fisher's z-transformed r values) across studies. Random-effects models assume that the true effect size varies between studies (not just due to sampling error), which is appropriate when studies differ in populations, settings, and measurement tools.

**Search strategy:** The authors searched six databases (Medline, CINAHL, Scopus, Cochrane Library, Web of Science, Sciencedirect) in December 2019, using PEO (Population, Exposure, Outcome) framework. They screened 139,552 initial hits, then 471 full-text papers, finally including 24 studies. Two reviewers independently screened papers; disagreements were resolved by a third reviewer.

**Quality assessment:** The authors used the **Risk of Bias in Non-randomised Studies (ROBINS-I)** tool from the Cochrane Handbook. This assesses bias across domains: confounding, selection bias, measurement of exposure, missing data, measurement of outcomes, and selective reporting. No study was excluded based on quality, but the authors note that all included studies were observational and therefore at risk of confounding.

**What this design can and cannot prove:**

**Can prove:** That there is a statistical association (correlation) between perceived stress and diet quality in this population. The meta-analysis gives a pooled estimate of the strength and direction of that association.

**Cannot prove:** That stress *causes* poor diet quality. Observational studies cannot establish causality because: (1) reverse causation is possible (poor diet might increase stress), (2) confounding variables (income, education, sleep, social support) could drive both stress and diet, (3) the cross-sectional design of most studies means stress and diet were measured at the same time, so temporal order is unknown.

**Cannot prove:** That reducing stress would improve diet quality. Only an RCT testing a stress-reduction intervention with diet as an outcome could address this.

**Major methodological weaknesses:**

1. **High heterogeneity:** I² = 93% — this is extremely high, meaning 93% of the variation between study results is due to real differences between studies, not chance. This makes the pooled estimate unreliable for predicting any individual's experience.

2. **Only 5 studies in the meta-analysis** — a very small number for a meta-analysis, limiting generalisability and statistical power for subgroup analyses.

3. **No biological stress measures** — all studies relied on self-reported perceived stress, which correlates only modestly with physiological stress markers like cortisol.

4. **Publication bias not formally assessed** — the authors note they could not test for publication bias due to the small number of studies.

5. **Cross-sectional designs dominate** — only 2 of the 8 diet quality studies were longitudinal; the rest measured stress and diet at one time point.

**Primary outcome — meta-analysis of diet quality studies (5 studies, n = 3,471):**

Pooled correlation between perceived stress and diet quality: **r = −0.35** (95% CI: −0.56 to −0.15, p < 0.001)

This is a **moderate negative correlation** — higher stress is associated with lower diet quality

Heterogeneity: **I² = 93%** (Cochran Q test p < 0.0001) — very high, meaning the true relationship varies substantially across populations and settings

After removing two outlier studies (sensitivity analysis), the correlation remained significant but the authors do not report the exact revised r value

**Secondary outcome — food intake studies (16 studies, not meta-analysed):**

Stress was consistently associated with **unhealthy dietary patterns**: higher intake of high-fat foods, sweets, salt, and fast food; lower intake of fruits, vegetables, fish, and unsaturated fats

These associations were statistically significant in most individual studies, but effect sizes were not pooled due to heterogeneity in outcome measures

**Individual study examples (from the paper's narrative synthesis):**

Barrington et al. (2012): Higher perceived stress was associated with higher fast food consumption in young women (β = 0.12, p < 0.05)

Habhab et al. (2009): Women in the high-stress group (unsolvable Sudoku) consumed more fats and sweets than the low-stress group (easy Sudoku) — but this was a small lab study (n = 40) with no baseline hunger control

Multiple studies: Women with higher PSS scores had lower fruit and vegetable intake (specific effect sizes not pooled)

**Notable null or contradictory findings:**

The paper notes that 40% of people eat more under stress, 40% eat less, and 20% eat the same — so the average effect masks substantial individual variation

Some studies found no significant association between stress and overall diet quality, only with specific food groups

The pooled correlation of r = −0.35 translates to a **moderate effect** in social science terms. To put it in plain English:

If you took two women from the same population, the one with higher perceived stress would, on average, have a diet quality score about **0.35 standard deviations lower** than the lower-stress woman

In practical terms, if diet quality is measured on a 0–100 scale (like the HEI), a one-standard-deviation increase in stress (roughly 6–8 points on the PSS-10) would be associated with a diet quality drop of about **5–8 points** — roughly the difference between a "needs improvement" and "good" diet rating

However, the 95% confidence interval is very wide (−0.56 to −0.15), meaning the true association could be anywhere from strong (r = −0.56, explaining ~31% of the variance) to weak (r = −0.15, explaining ~2% of the variance)

The high heterogeneity (I² = 93%) means that for some populations or individuals, the association might be much stronger or weaker — or even reversed

To give a concrete example from the food-intake studies: women in the highest stress quartile consumed approximately **1–2 fewer servings of fruits and vegetables per day** compared to women in the lowest stress quartile, and about **1–2 more servings of sweets or fast food per week**. These are meaningful differences — 2 fewer servings of vegetables per day is roughly the difference between meeting and not meeting dietary guidelines.

**What the authors acknowledge:**

1. **High heterogeneity** — the I² of 93% means the pooled estimate should be interpreted cautiously

2. **Only 5 studies in the meta-analysis** — limits statistical power and generalisability

3. **No biological stress markers** — all studies used self-report, which can be influenced by mood, memory, and social desirability

4. **Cross-sectional design of most studies** — cannot establish temporal order or causality

5. **Variety of diet quality indices** — different scoring systems make direct comparison difficult

6. **Publication bias** — could not be formally tested due to small number of studies

7. **English language only** — may have excluded relevant non-English studies

8. **Search conducted in December 2019** — studies published after this date are not included

**Additional limitations a critical reader would note:**

1. **Self-report bias for diet** — diet quality indices rely on self-reported food intake, which is notoriously inaccurate (people underreport calories by 20–40%, and overreport healthy foods)

2. **No control for key confounders** — most studies did not adequately control for income, education, sleep quality, physical activity, or social support, all of which affect both stress and diet

3. **Healthy volunteer bias** — women who volunteer for nutrition studies tend to be more health-conscious than the general population, potentially attenuating the stress-diet association

4. **No distinction between types of stress** — the PSS measures general perceived stress, but acute stress (deadline, exam) and chronic stress (poverty, caregiving) may have opposite effects on eating behaviour

5. **No measurement of emotional eating** — the stress-diet link is likely moderated by whether someone is an emotional eater, but none of the studies assessed this trait

6. **Industry funding not reported** — the paper does not disclose whether any included studies had food industry funding, which could bias results toward null findings

7. **Age range is broad** — women aged 18 and 49 have very different life circumstances, stress profiles, and dietary habits, but subgroup analyses by age were not possible

For someone running their own n=1 experiment:

**What to test:**

Test whether a specific stress-reduction technique (e.g., 10-minute daily mindfulness meditation, 20-minute brisk walk, or structured journaling) improves your diet quality

Alternatively, test whether a specific dietary change (e.g., adding one serving of vegetables to each meal) reduces your perceived stress

The paper suggests the relationship is bidirectional, so either direction is worth testing

**Minimum meaningful duration:**

Run the experiment for **at least 4 weeks** — the PSS asks about stress over the past month, and diet quality changes take 2–3 weeks to stabilise

For detecting a meaningful effect, **8–12 weeks** would be better, as stress-diet patterns may take time to shift

Measure daily for the first 2 weeks to establish baseline, then 4–8 weeks of intervention, then 2 weeks of follow-up

**What to measure (specific metrics):**

**Stress:** Use the **Perceived Stress Scale (PSS-10)** — free online, takes 5 minutes. Score 0–40. Measure weekly.

**Diet quality:** Use a simplified version of the **Healthy Eating Index** or **DASH score**. Track daily:

- Servings of fruits and vegetables (target: 5+ per day)

- Servings of whole grains (target: 3+ per day)

- Servings of sweets, sugary drinks, or fast food (target: 0–1 per day)

- Servings of fatty or fried foods (target: 0–1 per day)

**Optional biological marker:** Morning salivary cortisol (collect at waking, 30 min post-waking, and bedtime) — but this requires a lab kit and costs $50–100 per sample set

**Confounders to track daily:** Sleep quality (1–10 scale), physical activity (minutes), alcohol intake (drinks), menstrual cycle phase, work/school demands (1–10 scale)

**Key confounds to control for:**

1. **Sleep** — poor sleep increases stress and impairs dietary self-control. Track sleep duration and quality.

2. **Menstrual cycle** — stress perception and food cravings vary across the cycle. Track cycle phase (follicular vs. luteal).

3. **Alcohol** — alcohol is both a stress reliever and a source of empty calories. Track drinks separately from diet quality.

4. **Social support** — having someone to talk to buffers stress. Note days when you felt socially connected vs. isolated.

5. **Time of year** — seasonal affective changes, holidays, and exam periods all affect stress and diet. Run your experiment at a consistent time of year.

6. **Baseline diet** — if you already eat well, stress may have less room to worsen your diet. Start with a 1-week food diary to see your baseline.

**What a positive result would look like:**

A **decrease of 4–6 points on the PSS-10** (one standard deviation) over 4–8 weeks, combined with an **increase of 5–10 points on your diet quality score** (e.g., from 60 to 70 on a 0–100 HEI-like scale)

Or: a consistent pattern where on low-stress days (PSS < 12), you eat 1–2 more servings of vegetables and 0–1 fewer servings of sweets compared to high-stress days (PSS > 20)

The paper's meta-analysis