New methods for estimating follow-up rates in cohort studies

Table 1 Follow-up rates under varying assumptions estimated using four methods: (i) the standard Percentage Method (Eq. 1), (ii) the Clark’s Completeness Index (CCI, Eq. 2), (iii) the Person-Time Method estimated using the formal method (FPT, Eq. 4) and (iv) the Simplified Person-Time Method (SPT, Eq. 5)

Assumed event rate	True Person-time follow-up rate η _PTFR	Percentage Method η _percentage			Estimated using the formal method η _FPT			Clark’s compleness inex η _CCI			Simplified Person-time method η _SPT
Assumed event rate	True Person-time follow-up rate η _PTFR	Average	%bias¹	\( \sqrt{\mathrm{MSE}} \) ²	Average	%bias	\( \sqrt{\mathrm{MSE}} \)	Average	%bias	\( \sqrt{\mathrm{MSE}} \)	Average	%bias	\( \sqrt{\mathrm{MSE}} \)
5%	95.0%	90.4%	−4.90	.047	95.0%	0.00	.001	94.9%	−0.08	.001	95.1%	0.05	.001
	81.9%	66.7%	−18.5	.158	82.1%	0.16	.002	81.7%	−0.26	.003	82.1%	0.25	.003
	68.1%	44.8%	−34.3	.233	68.3%	0.45	.004	67.9%	−3.49	.003	68.2%	0.59	.005
	56.7%	29.3%	−48.2	.274	57.3%	0.92	.006	56.5%	−0.27	.003	57.3%	1.09	.007
10%	95.2%	91.0%	−4.48	.043	95.2%	−0.01	.002	95.1%	−0.29	.003	95.4%	0.10	.002
	82.1%	67.9%	−17.4	.142	82.3%	0.17	.002	81.6%	−0.69	.006	82.5%	0.49	.006
	68.5%	46.5%	−32.2	.220	68.9%	0.64	.005	67.9%	−1.11	.008	69.2%	1.07	.009
	56.4%	30.3%	−46.2	.261	57.1%	1.33	.008	55.6%	−1.36	.008	57.4%	1.84	.011
30%	94.4%	90.0%	−4.61	.044	93.4%	−0.95	.009	93.7%	−0.64	.006	94.6%	0.31	.004
	82.7%	70.7%	−14.5	.120	82.3%	−0.04	.004	81.1%	−1.94	.016	83.6%	1.06	.009
	69.4%	50.9%	−26.8	.186	69.7%	0.42	.005	67.1%	−3.33	.023	70.9%	2.17	.016
	53.6%	31.0%	−42.2	.226	54.8%	2.17	.012	51.1%	−4.71	.026	55.9%	4.19	.023
50%	93.2%	89.5%	−3.97	.037	88.3%	−0.53	.049	91.2%	−2.08	.020	93.9%	0.80	.008
	77.6%	67.2%	−13.4	.105	74.6%	−3.79	.030	72.5%	−6.58	.051	79.9%	3.00	.024
	65.0%	51.1%	−2.14	.140	63.6%	2.03	.014	58.5%	−9.93	.647	68.4%	5.25	.035
	46.6%	31.3%	−32.8	.153	47.8%	0.03	.014	40.0%	−14.0	.066	51.3%	10.0	.005

These results were compared to the true Person-time follow-up Rate (Eq. 3) based on complete information generated under the simulations, each averaged across 1000 simulated data sets. The simulations involved an assumed 5-year prospective cohort study of N = 1000 subjects with fixed annual interval clinical visits and non-informative dropout. Time-to-event was generated based on exponential distributions with event rates varied from 5 to 50% and time to dropout was generated based on an independent exponential distribution with dropout proportion varying from 10 to 70%. Results were averaged across the 1000 simulated datasets
Note: 1. % bias was calculated as (average of the particular method-η _PTFR)/η _PTFR*100%; 2. \( \sqrt{\mathrm{MSE}} \) was calculated as the square root of the average of (estimate-η _PTFR)². MSE from the true η _PTFRwas calculated instead of variance because several methods used here can be biased

ISSN: 1471-2288