North Carolina · 27 of 100 counties · exploratory

The Invisible PCOS Signal

PCOS has no dedicated public health surveillance system anywhere in the US. This checks whether its downstream marker — gestational diabetes — leaves a trace in county diabetes rates that poverty, obesity, and insurance access alone can't explain. The signal is real, weak, and not yet confirmed.

CDC PLACES · 2022 release Census ACS 5‑Year · 2022 CDC WONDER Natality · 2020–2024

27 / 100 Counties with resolvable GDM data WONDER suppresses counties <100k pop.

+0.021 R² Gain in explained variance from adding GDM rate p = 0.254 · not significant

r = 0.229 Correlation: "unexplained" diabetes × GDM rate Weak · positive

Onslow −2.8 High GDM, far lower diabetes than predicted Counter‑example

Where diabetes, poverty, and GDM overlap

Diabetes prevalence and poverty are resolvable for all 100 NC counties. GDM rate is only resolvable for the 27 counties large enough (population ≥100,000) to clear CDC WONDER's suppression threshold — the rest are shown in grey, not zero, since their true rate is simply unknown at this resolution.

Three choropleth maps of North Carolina counties: diabetes prevalence, poverty rate, and GDM rate (with most counties greyed out as unresolved due to CDC WONDER's small-county suppression rule). — Diabetes prevalence, poverty rate, and 2020–2024 mean GDM rate by county. Sources: CDC PLACES 2022 release; U.S. Census Bureau ACS 5-Year 2022; CDC WONDER Natality.

Does GDM explain the leftover diabetes burden?

Two models, same 27-county subsample so the comparison is apples-to-apples: a baseline of poverty, obesity, and uninsured rate — the "already well-researched" story — then GDM rate added on top.

diabetes_prev ~ poverty_rate + obesity_prev + no_insurance_prev + gdm_rate

Term	Coef.	p‑value	Significant
Intercept	−0.47	0.876	No
Poverty rate	+0.103	0.314	No
Obesity rate	+0.243	0.049	Yes
Uninsured rate	+0.083	0.728	No
GDM rate	+0.170	0.254	No

R² goes from 0.643 to 0.664 when GDM rate is added — a small gain, and GDM's own coefficient (+0.17) doesn't clear conventional significance (p = 0.254) at n = 27.

Of the four predictors, only obesity rate reaches significance (p = 0.049) — consistent with established literature. Poverty, insurance access, and GDM rate all point the expected direction but are individually underpowered at this sample size.

This is exactly the shape of result a genuine-but-modest effect would produce at n = 27 — and also exactly the shape pure noise would produce. The data alone can't tell those apart yet.

Which counties carry the signal — and which don't

Residual = actual diabetes prevalence minus what the poverty/obesity/insurance model alone predicts. Positive (rust) means "higher than expected"; negative (blue) means "lower than expected" — plotted here against each county's GDM rate.

Map of the 27 GDM-resolvable NC counties shaded by diabetes residual. Robeson and Brunswick counties show the strongest positive (rust) residuals; Onslow shows the strongest negative (blue) residual. — Residual hotspot map — only the 27 GDM-resolvable counties are colored.

Scatter plot of diabetes residual against GDM rate for 27 NC counties, with a weak positive trend line. Brunswick, Henderson, and Robeson sit well above the line; Onslow sits far below it as a clear counter-example. — Residual vs. GDM rate, all 27 counties labeled. The trend is weak and driven by a handful of points.

County	Diabetes %	Poverty %	Obesity %	GDM %	Residual
Brunswick	13.5	9.1	32.8	10.2	+3.01
Henderson	12.5	11.4	30.3	10.8	+2.05
Robeson	18.4	27.1	47.9	7.0	+1.86
Wayne	14.2	17.9	39.5	9.5	+0.77
Davidson	12.7	13.9	36.2	5.7	+0.61
Gaston	12.3	12.4	36.2	9.9	+0.40
Buncombe	10.9	11.2	31.3	8.7	+0.36
Craven	12.5	14.0	36.9	9.3	+0.36
Johnston	11.6	10.7	35.1	7.0	+0.05
Cabarrus	10.8	8.4	33.3	10.9	+0.05
Iredell	10.7	9.4	32.8	8.1	+0.04
Randolph	13.2	14.7	39.9	9.7	+0.03
Rowan	12.7	16.4	37.4	8.5	+0.02
New Hanover	10.6	12.7	31.1	8.6	+0.02
Union	10.5	6.8	33.4	9.4	−0.03
Catawba	12.1	13.0	37.4	8.1	−0.12
Mecklenburg	10.1	10.5	29.7	8.5	−0.16
Wake	9.5	8.2	29.3	8.5	−0.16
Guilford	12.2	15.1	38.0	8.2	−0.30
Alamance	12.2	14.2	38.5	5.0	−0.47
Orange	9.3	12.3	28.0	7.6	−0.49
Harnett	11.8	14.4	36.9	7.7	−0.60
Forsyth	12.4	15.2	41.2	4.6	−0.84
Cumberland	12.2	17.6	40.1	8.8	−1.17
Durham	10.5	12.3	36.1	5.7	−1.22
Pitt	12.0	20.0	39.0	10.3	−1.23
Onslow	8.6	12.8	33.4	9.6	−2.85

🔍 Before you read this as a finding

This is hypothesis-generating, not confirmed. At n = 27 — the ceiling CDC WONDER's county-suppression rules allow — a weak positive relationship is compatible with both "a real, modest effect" and "no effect, just noise."
The relationship isn't visually robust. Brunswick, Henderson, and Robeson are doing most of the work pulling the trend line upward. Onslow is a strong negative outlier — a high GDM rate paired with much lower diabetes than predicted — that a purely narrative reading would conveniently ignore.
County-level correlation isn't individual-level linkage. This can't show that the same women who had GDM are the same ones later diagnosed with type 2 diabetes — that's the ecological fallacy, and it's a structural limit of any analysis at this resolution.
What would strengthen it: more counties (would require a restricted-access data-use agreement with NC DHHS to get past WONDER's 100k-population floor), and a true individual-level GDM→later-T2D linkage instead of a county-level comparison.