Cohort · india-sim

simulated at Indian frequenciesDownload CSV

Group-level statistics across de-identified records. Each record came from a lab report, a WhatsApp photo, dictated results or a VCF, read by Muse Spark and standardised by PharmCAT. Research output, not medical advice.

120

Records

120

From reports and photos

Curator verified

Variants with coverage gaps

Simulated people, real sampling question. Each record is a synthetic report for a person whose genotype was drawn at published Indian allele frequencies, issued by a lab that tests only part of the panel. Because the complete-panel answer is known, both ways of pooling can be scored against it. Neither column is the truth: the left one assumes a report that never named a variant tells you nothing about it, the right one assumes it means absent. The gap between them is the uncertainty a cohort of partially-covering panels actually carries, and chapter 04 reports what each one cost.

Variant	Carriers	Carrier frequency, 95% CI	Left out of the denominator	If counted as absent	Simulated frequency
DPYD HapB3	2/34	5.9% (1.6%–19%)	4516 undocumented	2.5%	3.4%
DPYD c.85T>C (*9A)	12/39	31% (19%–46%)	4013 undocumented	15%	24%
DPYD *2A	0/61	0.0% (0.0%–5.9%)	1816 undocumented	0.0%	0.2%
DPYD *13	0/61	0.0% (0.0%–5.9%)	1816 undocumented	0.0%	0.2%
DPYD c.2846A>T	1/61	1.6% (0.3%–8.7%)	1816 undocumented	1.3%	0.4%
NUDT15 *3	9/29	31% (17%–49%)	1010 undocumented	23%	17%
UGT1A1 *28	44/70 · 11 hom	63% (51%–73%)	98 undocumented	56%	58%
UGT1A1 *6	1/34	2.9% (0.5%–15%)	4516 undocumented	1.3%	4.0%
TPMT *3C	4/29	14% (5.5%–31%)	1010 undocumented	10%	2.8%
CYP2C19 *2	24/37 · 2 hom	65% (49%–78%)	55 undocumented	57%	58%
CYP2C19 *17	14/34 · 1 hom	41% (26%–58%)	88 undocumented	33%	28%

Carriers have at least one copy. A record whose panel did not name a variant is left out of that variant's denominator instead of being counted as a non-carrier: either the report lists its panel and the variant is not in it, or the report never says which variants it tested, which is just as unknowable and is counted separately. Calls a curator marked unreadable are excluded. Intervals are Wilson 95% intervals. “If counted as absent” is the same records pooled the common way, with every panel that never tested a variant counted as a non-carrier. The simulated frequency is what each person's genotype was drawn from; it is a simulation input, not a measurement.

9 of 11 variants have the frequency they were drawn from inside the coverage-aware 95% interval, and leaving untested panels out moves the estimate up by 5.0 points on average across the 11 variants with a coverage gap. Scored against the complete-panel answer, the naive column was actually closer on average here (1.3 against 4.1 points): leaving records out is unbiased but noisy at this size. The column that matters is DPYD c.85T>C, the variant a European tier-1 panel never tests: 15% counted absent, 31% left out, against a true 25%. Full numbers in docs/pooling-results.json.