Can we quantify harm in general practice records? An assessment of precision and power using computer simulation

Table 2 The precision ^$ of selected clinical record review (CRR) scenarios’ estimated harm rates

Practices (n)	Records reviewed per practice (n)	Estimation error (%)^$
		rHR^* = 2	rHR^* = 5	rHR^* = 10	rHR^* = 20
1	20	-	-	-	-
	25	-	-	-	-
	50	-	-	-	-
	100	-	-	-	107
	150	-	-	132	88
	200	-	-	107	74
10	20	-	-	107	74
	25	-	146	96	67
	50	-	96	65	46
	100	107	65	46	32
	150	88	54	37	26
	200	74	46	32	23**
20	20	-	107	74	51
	25	-	96	65	46
	50	107	65	46	32
	100	74	46	32	23
	150	60	37	26	18
	200	51	32	23	16
50	20	107	67	46	32
	25	96	58	41	29
	50	65	41	29	20
	100	46	29	20	14
	150	37	23	17	12
	200	32	20	14	10
100	20	74	46	32	23
	25	65	41	29	20
	50	46	29	20	14
	100	32	20	14	10
	150	26	17	12	8
	200	23	14	10	7
150	20	60	37	26	18
	25	54	33	23	17
	50	37	23	16	12
	100	26	17	12	8
	150	21	13	10	7
	200	18	12	8	6
200	20	52	32	23	16
	25	46	29	20	14
	50	32	20	14	10
	100	23	14	10	7
	150	18	12	8	6
	200	16	10	7	5
250	20	46	29	20	14
	25	41	26	18	13
	50	29	18	13	9
	100	20	13	9	6
	150	17	10	7	5
	200	14	9	6	5
300	20	42	26	18	13
	25	37	23	17	12
	50	26	17	12	8
	100	18	12	8	6
	150	15	10	7	5
	200	13	8	6	4

^$Precision is expressed as percentage estimation error, averaged across 1000 simulated studies for each CRR scenario.
*rHR (real harm rate): The actual, underlying ‘baseline’ harm rate, expressed as number of incidents/100 patients/year. The harm rates estimated by different CRR scenarios are not shown.
**Example: In this CRR scenario, 10 practices each reviewed 200 records and the estimation error was ±23%, e.g. within ±23% of the rHR of 20 incidents/100patients/year. The estimation error (%) indicates the proximity between the harm rate estimated by that unique CRR scenario and the rHR. Smaller estimation errors therefore indicate greater precision. We defined acceptable precision as estimation errors < ±25%.
Scenarios vary by numbers of practices reviewing records, number of records reviewed in each practice and real harm rates (rHR)*. The median rate ratios (MRR) between patients and practices are 2 and 1.2 respectively. The results are from the beginning of the simulated 12-month period.