SSRI STATISTICAL METHODS AND RESULTS 6 November 2026 - review edition 6 Core area summaries and exploratory tests were calculated on 4 November 2026. This edition adds the visual decomposition and Lothian contribution check using the same existing records. Tests remain in the expandable methods. QUESTION AND UNIT Describe the size, area variation and temporal consistency of the 2024-2025 changes in community SSRI dispensing. All calculations use the existing 24-month reconciled extract. Areas are dispensing locations, not patient residences or prescriber exposure groups. No individual records, diagnoses, new starts, switches or outcomes are available. DESCRIPTIVE EFFECT SIZES Sertraline: national item change +3.3651%; 2025/2024 count ratio 1.033651. Area median +2.7763%; IQR +2.3314% to +3.8598%; range +1.0133% to +9.3390%. All 14 areas have positive 2024 denominators. Citalopram: national item change -3.8404%; 2025/2024 count ratio 0.961596. Area median -3.5324%; IQR -3.9389% to -2.7710%; range -6.8109% to -0.4712%. All 14 areas have positive 2024 denominators. Each area's percentage is 100 x (2025 items / 2024 items - 1). The national percentage uses national sums, not the average of the 14 area percentages. Area medians and quartiles give each area equal weight. Quartiles use linear interpolation at index (n-1) x p (zero-based), with p=.25 or .75. These IQRs describe the observed middle 50% of area changes, not uncertainty around a mean or patient outcome. CONSISTENCY AND SENSITIVITY For each medicine, compare its national share of all six SSRI items for January 2025 against January 2024; repeat for all 12 months. Sertraline's share is higher in 12/12 pairs; citalopram's is lower in 12/12. This is not a claim that both medicines' raw item counts moved the same way in every month. These month pairs are serially related; no monthly t-test or binomial p-value is attached. For each of 14 areas, omit January from both years and compare annual medicine counts; repeat all months. Sertraline remains positive in 168/168 checks. Citalopram remains negative in 167/168. The exception is Dumfries and Galloway when May is excluded: +15 items, compared with the full-year -150. These overlapping sensitivity checks are not 168 independent observations. Divide 2024 counts by 366 days and 2025 by 365. Sertraline and citalopram retain their respective up/down directions in all 14 areas. National per-calendar-day changes are +3.6483% and -3.5770%. This adjustment does not standardise working days, payment timing, population or clinical need. EXPLORATORY DIRECTION TESTS One observation per area: the sign of its annual count difference. All six SSRI medicines are shown, not only the two unanimous results. A two-sided exact binomial sign test uses H0: P(increase | nonzero change)=0.5, with independent area signs. Zero differences are excluded; Fluvoxamine has two ties, so its test n=12 rather than 14. Other test n=14. Medicine | Up | Down | Tied | Raw p | Holm-adjusted p Sertraline | 14 | 0 | 0 | 0.000122070 | 0.000732422 Citalopram | 0 | 14 | 0 | 0.000122070 | 0.000732422 Fluoxetine | 8 | 6 | 0 | 0.790527344 | 1.000000000 Escitalopram | 12 | 2 | 0 | 0.012939453 | 0.051757812 Paroxetine | 2 | 12 | 0 | 0.012939453 | 0.051757812 Fluvoxamine | 7 | 5 | 2 | 0.774414062 | 1.000000000 At p=.5, raw two-sided p = min(1, 2 x sum[choose(n,k)/2^n, k=0..min(up,down)]). For all 14 changes in one direction, 2/2^14 = 0.0001220703125. The calculation was independently checked by enumerating all 2^n possible sign vectors for every medicine. Holm correction sorts the six raw p-values; multiply ordered p[i] by (6-i), take the cumulative maximum and cap at 1. Both unanimous medicines have corrected p=0.000732421875. The two 12-versus-2 medicines have corrected p=0.0517578125; the remaining two have corrected p=1. No binary significant/not-significant labels are applied. WHAT THESE P-VALUES CAN SUPPORT They quantify incompatibility with the stated independent, balanced-sign model. They are not the probability that a hypothesis is true or the data arose by chance, and do not measure effect size or clinical importance. This analysis was developed after the data and story had been examined, not preregistered. Correction covers these six SSRI tests only, not the project's earlier exploratory searches. The 14 areas are the full observed territorial set, not a random sample. They share guidance, policies and wider influences; spatial/health-system dependence can invalidate the independent-sign model. Holm can handle dependence BETWEEN otherwise valid tests, but does not fix dependence WITHIN an area-sign test. These results are conditional exploratory calculations, not calibrated confirmatory evidence of a national mechanism. Independence sensitivity illustration: seven hypothetical independent signs all agreeing give raw p=2/2^7=0.015625; four give 2/2^4=0.125. These are not estimated effective sample sizes, alternative Scottish area groupings or adjusted p-values. They illustrate the importance of the assumption. WHY NO ITEM-LEVEL CONFIDENCE INTERVALS The observed totals, area median and empirical spread are known for the included records and period. Repeated dispensing items cannot be treated as independent patient outcomes. A confidence or prediction interval for another period, clinical effect or underlying process requires an explicit justified sampling/process model and dependence structure. We do not attach binomial/Poisson sampling intervals or item-level chi-squared tests to these complete recorded counts. Data quality and coverage uncertainty are not eliminated by observing the full administrative extract. USE IN THE STORY Lead with effect sizes and observed consistency. Keep the p-values alongside their model, all six results and limitations. They do not establish patient switching, treatment benefit, worsening mental health, guidance effects, cost-driven decisions or unwarranted area variation. Reader/clinical review is still pending. Nothing has been submitted; reader feedback review and explicit user readiness remain required. METHOD REFERENCES Exact binomial tests: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.binomtest.html Holm (1979), implementation reference: https://stat.ethz.ch/R-manual/R-devel/library/stats/html/p.adjust.html ASA principles: https://www.amstat.org/asa/files/pdfs/p-valuestatement.pdf Data: https://www.opendata.nhs.scot/dataset/prescriptions-in-the-community LOTHIAN CONTRIBUTION CHECK — 6 NOVEMBER 2026 Lothian's sertraline items: 229,831 (2024) to 251,295 (2025), +21,464. Scotland: 2,066,436 to 2,135,974, +69,538. Baseline share = 100 x 229,831 / 2,066,436 = 11.1220962% (11.1% displayed). Share of increase = 100 x 21,464 / 69,538 = 30.8665765% (30.9% displayed). These shares have different denominators. They do not measure population, clinical need, patient switching or quality of care. The national increase remains positive outside Lothian; the pattern is widespread, with uneven contributions. Waterfall running totals: 0 + 69,538 = 69,538; minus 37,514 = 32,024; minus 553 = 31,471. Final bar independently starts at zero. All bars use the same 0-to-80,000 item scale, with no minimum-height inflation of the small 553-item change.