Data were available from N=5,048 participants with migraine in the population-based sample and from N=2,752 in the Eurolight sample. Not all provided a complete set of responses required for these analyses: an account of missing data is in Table 1. In particular, among the population-based sample, one third (35.5%) of males did not provide responses for lost household days and one quarter (25.3%) of females did not do so for lost work days. These gender-based differences were not seen among the Eurolight sample, whose responder proportions were invariably higher (Table 1).
Descriptives
The regression analyses were performed on the IPD, but the data on lost productivity and attack frequency, duration and intensity are summarised in Table 2, stratified by gender and sample. Overall, medians were lower than means, indicating skewedness in the data. All SEMs were small, indicating that sample means were accurate estimates of the true population means. Females had migraine episodes more frequently than males, and, in the Eurolight sample, duration was longer for females than males. Intensity was similarly distributed between genders, but the proportions with “very bad” headache were greater in the population-based sample than in the Eurolight sample. Lost productivity in paid work was similar between males and females in the Eurolight sample (2.5 vs 2.7 days/3 months), whereas females in the population-based sample lost fewer days from paid work (2.2 days vs 3.3 days/3 months). Females in both samples reported greater losses than males from household chores (3.9-4.3 vs 2.8-3.0 days/3 months).
Table 1. Numbers of participants and of those with missing data for each variable
|
Population-based
|
Eurolight
|
Male
|
Female
|
Male
|
Female
|
N with migraine
|
1,856
|
3,192
|
870
|
1,882
|
Numbers missing data
|
Lost productivity
HALT questions 1+2
HALT questions 3+4
|
136
578
|
664
310
|
42
46
|
172
115
|
Disability factors
F only
D only
I only
F+D
F+I
D+I
F+D+I
Total with missing disability data
of whom also missing HALT 1+2 data*
and of whom also missing HALT 3+4 data*
|
17
88
4
5
0
2
0
116
11
35
|
10
156
5
10
0
1
0
182
37
32
|
2
5
3
1
0
3
0
14
2
2
|
2
26
17
4
1
10
1
61
18
14
|
Final N in HALT 1+2 analyses*
|
1,615 (87.0%)
(1,856 – [136+(116-11)])
|
2,383 (74.7%)
(3,192 – [664+(182-37)])
|
816 (93.8%)
(870 – [42+(14-2)])
|
1,667 (88.6%)
(1,882 – [172+(61-18)])
|
Final N in HALT 3+4 analyses*
|
1,197 (64.5%)
(1,856 – [578+(116-35)])
|
2,732 (85.6%)
(3,192 – [310+(182-32)])
|
812 (93.3%)
(870 – [46+(14-2)])
|
1,720 (91.4%)
(1,882 – [115+(61-14)])
|
HALT: Headache-Attributed Lost Time, F: headache frequency, D: headache duration, I: headache intensity, *corrections applied to avoid double counting
Table 2. Symptom burden (attack frequency, duration and intensity) and lost productivity (work days [HALT 1+2] and household days [HALT 3+4]) in the two samples
Sample
|
Frequency (days/month)
|
Duration (hours)
|
Intensity
|
HALT 1+2 (days/3 months)
|
HALT 3+4 (days/3 months)
|
mean ± SEM (median)
|
Not bad n (%)
|
Quite bad n (%)
|
Very bad n (%)
|
Mean ± SEM (median)
|
Population-based
|
male
|
2.7 ± 0.1 (2.0)
|
24.9 ± 1.2 (6.0)
|
104 (5.6)
|
1012 (54.7)
|
734 (39.7)
|
3.3 ± 0.2 (1.0)
|
3.0 ± 0.2 (0.0)
|
female
|
3.2 ± 0.1 (2.0)
|
26.7 ± 0.9 (12.0)
|
162 (5.1)
|
1682 (52.8)
|
1342 (42.1)
|
2.2 ± 0.1 (0.0)
|
3.9 ± 0.2 (2.0)
|
Eurolight
|
male
|
2.5 ± 0.1 (1.7)
|
20.9 ± 1.1 (8.0)
|
164 (19.0)
|
522 (60.4)
|
178 (20.6)
|
2.5 ± 0.3 (0.0)
|
2.8 ± 0.3 (0.0)
|
female
|
3.3 ± 0.1 (2.5)
|
37.3 ± 1.1 (24.0)
|
179 (9.7)
|
1138 (61.4)
|
536 (28.9)
|
2.7 ± 0.1 (0.0)
|
4.3 ± 0.2 (2.0)
|
HALT: Headache-Attributed Lost Time, SEM: standard error of mean
Table 3. Multiple linear regressions predicting lost productivity (work days and household days) from frequency, duration and intensity of migraine attacks
Sample
|
N
|
Regression model
|
Equation (unstandardized coefficients)
|
Standardized coefficients
|
VIF
|
p
|
F
|
D
|
I
|
F
|
D
|
I
|
HALT questions 1+2 (lost work days per 3 months)
|
Population-based
|
male
|
1,615
|
F (3, 1611) = 94.3 p<0.001, R2=0.15
|
Y = 0.85xF + 0.01xD + 1.30xI – 0.93
|
0.37
|
0.03
|
0.11
|
<1.04
|
<0.001
|
0.17
|
<0.001
|
female
|
2,383
|
F (3, 2379) = 38.2 p<0.001, R2=0.05
|
Y = 0.34xF + 0.01xD + 0.91xI – 1.05
|
0.18
|
0.06
|
0.10
|
<1.03
|
<0.001
|
0.007
|
<0.001
|
Eurolight
|
male
|
816
|
F (3, 812) = 27.8 p<0.001, R2=0.09
|
Y = 0.75xF + 0.02xD + 1.32xI – 2.46
|
0.26
|
0.06
|
0.10
|
<1.05
|
<0.001
|
0.06
|
0.003
|
female
|
1,667
|
F (3, 1663) = 63.0, p<0.001 R2=0.10
|
Y = 0.53xF + 0.01xD + 1.31xI – 2.14
|
0.26
|
0.04
|
0.13
|
<1.07
|
<0.001
|
0.14
|
<0.001
|
|
HALT questions 3+4 (lost household days per 3 months)
|
Population-based
|
male
|
1,197
|
F (3, 1193) = 34.9 p<0.001, R2=0.08
|
Y = 0.67xF + 0.01xD + 1.15xI – 1.77
|
0.25
|
0.07
|
0.10
|
<1.05
|
<0.001
|
0.01
|
0.001
|
female
|
2,732
|
F (3, 2728) = 122.2 p<0.001, R2=0.12
|
Y = 0.89xF + 0.01xD + 1.43xI – 0.98
|
0.33
|
0.03
|
0.10
|
<1.03
|
<0.001
|
0.14
|
<0.001
|
Eurolight
|
male
|
812
|
F (3, 808) = 52.5 p<0.001, R2=0.16
|
Y = 0.87xF + 0.04xD + 1.31xI – 2.91
|
0.31
|
0.16
|
0.11
|
<1.05
|
<0.001
|
<0.001
|
0.001
|
female
|
1,720
|
F (3, 1716) = 112.2 p<0.001, R2=0.16
|
Y = 0.83xF + 0.02xD + 2.07xI – 3.53
|
0.32
|
0.10
|
0.16
|
<1.08
|
<0.001
|
<0.001
|
<0.001
|
F: frequency of migraine attacks (continuous, measured in days/month), D: duration of migraine attacks (continuous, measured in hours), I: intensity of migraine attacks (ordinal: 1 = “not bad”, 2 = “quite bad”, 3 = “very bad], VIF: variance inflation factor
Graphic visualizations
In Figures 1-3, frequency, duration and intensity of migraine attacks are plotted against lost productivity in paid work (lost work days) and household chores (lost household days). No direct statistical tests were performed, but the visualizations clearly show positive linear relationships between frequency and intensity on the one hand and lost productivity on the other in all groups. Duration had no such relationship: attacks reportedly lasting from two to 24 hours were associated with very similar productivity losses, with small up-kicks at the far-right indicative of impacted productivity on the next day from attacks of >24 hours’ duration. In the population-based sample, headache had greater impact on productivity in paid work in males than in females, and the opposite in household chores. Gender differences were small or none in the Eurolight sample.
[Figures 1-3 near here]
Multiple linear regressions
Multiple linear regressions were performed on the IPD to predict lost productivity from attack frequency, duration and intensity (Table 3). Values of R2 were small, and ranged from 0.05 to 0.16 because of high variance, but all regression models were highly significant (p<0.001). Therefore, it was possible to use the equations to predict productivity losses at population level. The VIFs were small (<1.08), indicating no collinearity between the predictors.
Lost productivity in paid work
Both frequency and intensity of migraine attacks were significant predictors of lost productivity in paid work in males, whereas duration was not: in the population-based sample, predicted productivity in paid work decreased by 0.85 days/3 months for each marginal increase of 1 headache day/month and by 1.3 days/3 months for each one-step increment in intensity, but by only 0.01 days/3 months for each marginal increase of 1 hour in duration (unstandardized coefficients: Table 3). Findings were similar in the Eurolight sample. The standardized regression coefficients showed that frequency was a much better predictor of lost productivity in paid work than intensity or duration (Table 3).
There were some gender-related differences. While results were mostly similar between males and females in the Eurolight sample, each marginal increase of 1 headache day/month led to a slightly greater decrease in productivity in males than in females (0.75 vs 0.53 days/3 months). In the population-based sample, frequency was a much more important predictor of lost productivity in males than in females (0.85 vs 0.34 days/3 months). Furthermore, duration was a significant predictor for lost productivity in paid work in females but not in males.
Lost productivity in household chores
As in paid work, the standardized regression coefficients demonstrated frequency to be the best predictor by far of lost productivity in household chores in both genders (Table 3). Frequency, duration and intensity of migraine attacks were all significant predictors of lost productivity in household chores in males. This was also true for females in the Eurolight sample, whereas only frequency and intensity were significant among females in the population-based sample. Similar regression coefficients for frequency were found for both genders in the Eurolight sample and for females in the population-based sample: each marginal increase of 1 headache day/month led to decreased productivity in the range of 0.83-0.89 days/3 months. Impact of frequency was somewhat less among males in the population-based sample (0.67 days/3 months).
Overall, the standardized and unstandardized regression coefficients for duration and intensity were quite similar between the different regression equations (Table 3). Frequency on the other hand, was more important in predicting household losses than those from paid work among females, whereas the opposite was true for males in the population-based sample.