Monitoring patient care through health facility exit interviews: an assessment of the Hawthorne effect in a trial of adherence to malaria treatment guidelines in Tanzania

Table 2 Hawthorne effect on data recording and malaria practice

	Non-survey days		Survey days		Adjusted comparison^a
Number of consultations per day					Difference	95 % CI	p
Mean (SD)	11.9	(7.3)	14.1	(10.3)	2.03	1.20- 2.86	<0.001
Median (IQR)	11	(7–16)	12	(8–17)
Missing MTUHA information	n	%	n	%	OR	95 % CI	p
Age	133	1.4 %	106	1.1 %	0.65	0.46-0.91	0.011
Gender	36	0.4 %	13	0.1 %	0.20^b	0.05-0.86^b	0.031^b
Village of origin	5,937	60.4 %	6,919	71.0 %	1.65^c	1.13-2.40^c	0.010^c
Previous attendance	1,604	16.3 %	2,127	21.8 %	0.54	0.26-1.13	0.103
Subscriber type	4,152	42.2 %	5,785	59.4 %	1.92^c	1.38-2.67^c	<0.001^c
Malaria diagnostic and treatment (primary outcomes)	n/N	%	n/N	%	OR	95 % CI	p
RDT result recorded	1,811/9,834	18.4 %	2,010/9,745	20.6 %	1.11	0.98-1.26	0.097
AM prescription with a negative RDT	168/1,721	9.8 %	160/1,901	8.4 %	0.83	0.56-1.23	0.343
AM prescription without a RDT result	210/8,023	2.6 %	182/7,735	2.4 %	0.73	0.53-1.00	0.052

^aComparison of survey days to non-survey days, from mixed-effect logistic regression, adjusted for day of the week (Monday-Friday) and study period. Analyses based on three-level hierarchical models (with health facility and day of data collection as random effects), except for number of consultations per day, based on a two-level hierarchical model (health facility as random effect)
^bUnadjusted, due to sparse data
^cOne-level logistic model, with robust standard errors for health facility clustering, due convergence failure for the hierarchical model
RDT Rapid Diagnostic Test, AM Antimalarial treatment, SD Standard deviation, OR Odds Ratio, CI Confidence Interval

ISSN: 1471-2334