Effects of neck-exercise and health promotion on headache outcomes in office workers: secondary analysis of the NEXpro stepped wedge cluster randomised controlled trial

Table 3 Poisson models for headache frequency

Outcome Variable: Headache days in the past four weeks
Poisson model: Incidence of headache events in the last four weeks	Model 1 ~ Intervention	Model 2 ~ Intervention + Period	Model 3 ~ Intervention x Period	Model 4 ~ Period + Exposure time (Intervention nested in exposure time)
Independent variable	Estimates (95%CI)	Estimates (95%CI)	Estimates (95%CI)	Estimates (95%CI)
Constant	1.97 (1.58 to 2.47)	2.04 1.61 to 2.57)	2.03 (1.61 to 2.57)	2.04 (1.62 to 2.58)
Intervention	0.84 (0.75 to 0.94)	0.78 (0.65 to 0.93)		-
Period 1		0.99 0.85 to 1.15)	1.08 (0.92 to 1.26)	0.96 (0.83 to 1.12)
Period 2		0.88 (0.75 to 1.04)	1.13 (0.95 to 1.34)	0.84 (0.71 to 1.00)
Period 3		1.03 0.85 to 1.24)	0.83 (0.65 to 1.05)	0.85 (0.69 to 1.07)
Period 4		1.12 (0.89 to 1.42)	0.87 (0.75 to 1.01)	0.90 (0.66 to 1.23)
Intervention x Period 1			0.57 (0.44 to 0.74)
Intervention x Period 2			0.77 (0.58 to 1.01)
Intervention x Period 3			1.07 (0.81 to 1.40)
Exposure: 4 months				0.82 (0.67 to 1.00)
Exposure: 8 months				0.88 (0.67 to 1.14)
Exposure: 12 months				1.20 (0.87 to 1.67)
Exposure: 16 months				1.19 (0.78 to 1.81)
Observations	519	519	519	519
Log likelihood	−1091	−1088	−1082	−1083
Akaike Information Criterion (AIC)	2191	2192	2184	2184
Bayesian Information Criterion (BIC)	2208	2226	2226	2235

Estimates are incidence rate ratios (IRR) for headache events in the last four weeks, with 95% confidence intervals. Figures in bold indicate the best fitting Model
Model one demonstrated an IRR of 0.84 (95% CI: 0.75 to 0.94) in favour of the intervention
Model two indicated no period effect. A period-averaged IRR of 0.78 (95% CI: 0.65 to 0.93) in favour of the intervention was found
Model three showed a significant interaction between intervention and period, but no main effect for period (see also Fig. 3) After having received the intervention in period one, an IRR of 0.57 (95% CI: 0.44 to 0.74) in favour of the intervention was demonstrated
ANOVA showed a better statistical fit for Model three compared to Model one (p= 0.0021), also indicated by the AIC
Model four indicated a significant effect for period two with an IRR of 0.84 (95% CI: 0.71 to 1.00). The intervention nested in exposure time indicated better intervention effects after shorter periods, especially four months with an IRR of 0.82 (0.67 to 1.00)

ISSN: 1129-2377