The project was approved by the Wellington regional ethics committee (reference 03/02/004). Participation was voluntary and anonymous. Participants had the option to enter a prize draw for a 2‐day mystery break for two adults (airfare, meals and accommodation within New Zealand).
A four page questionnaire (available from the corresponding author) was developed. An initial version was piloted in a survey of 192 junior doctors in two local hospitals (50% response rate), as a public health research project by a group of fifth year medical students. On the basis of this project, the questionnaire and survey distribution methods were revised. The revised questionnaire was then completed by a further group of junior doctors in an informal focus group, where their comments and suggestions were elicited. The final questionnaire sought information on demographics and professional activities; work patterns in each of the previous 2 weeks20
; sleep habits21
and sleepiness (in daily life in general22
and while driving23
); fatigue‐related errors in clinical practice24
; availability of different types of support, including shift work education, supervision, and access to napping facilities, healthy food, and personal counselling;20
and effects of work patterns on aspects of life outside work.25
Space was also provided for written comments.
Study packages contained a cover letter, a copy of the questionnaire, a work sheet to aid recall of work patterns, and a prepaid return envelope addressed to the Sleep/Wake Research Centre. In August 2003, the Resident Doctors' Association (the junior doctors' union, the RDA) mailed study packages to all junior doctors (not only its members) nation wide. (When they start employment, junior doctors' contact details are forwarded to the RDA so that the Association can mail out invitations to take up membership). This process was used to ensure confidentiality, since the Sleep/Wake Research Centre had no knowledge of names or addresses, and the RDA had no access to completed questionnaires. The only shared information was a unique study ID number for each participant. Non‐responders were sent a reminder postcard 3 weeks later, followed by a new study package 4 weeks after the postcard, with all mail‐outs conducted through the RDA. Participants were offered the opportunity to have their ID number entered into a prize draw for a commercially available mystery weekend for two people (airfares to a destination disclosed on arrival at the airport, with accommodation and meals).
In December 2003, feedback indicated that some junior doctors had not received study packages, probably as a result of having moved between hospitals. New packages were therefore distributed directly through unit managers responsible for junior doctors in every public hospital (more than 99% of junior doctors work in public hospitals). Participants who had already completed the questionnaire were asked not to respond a second time, but this could not be verified.
The distribution process meant that we could not use the number of packages mailed as the denominator for calculating response rates. We therefore used as the denominator the latest available workforce data from the Medical Council of New Zealand, with whom all medical practitioners must register annually.
All questionnaires were double entered. Statistical analyses were conducted using SAS Version 8.02 (SAS Institute, Inc Cary, NC, USA) and the Statistics Package for Social Sciences (SPSS), version 12.0.1 (SPSS Incorporated Software, Chicago, Illinois, USA).
Each participant was assigned a total risk score according to his/her work patterns in the preceding week, based on the risk assessment guide (page 14) in the AMA Code.20
The only risk factor not included in the present study was the direction of shift rotation, because it is difficult to define in some schedules, particularly where there is on‐call work. Table 1 summarises the scoring system (maximum possible score
20). For logistic regression analyses, total risk scores were divided into three categories (terciles): “lower risk”, scores of 0–6 (37% of participants); “significant risk”, scores of 7–9 (31% of participants); and “higher risk”, scores of 10–20 (32% of participants).
Table 1Calculation of total risk scores
Logistic regression analysis was used to identify independent risk factors for excessive sleepiness (Epworth Sleepiness Score (ESS) >10); feeling close to falling asleep at the wheel in the past 12 months; and recalling fatigue‐related error(s) in clinical practice in the past 6 months. The independent variables considered were gender; age (in quartiles); ethnicity (Maori, non‐Maori New Zealander, or of other ethnicity)26
; whether participants had dependents living with them; commute time (1–15 minutes, 16–30 minutes, >30 minutes); how often participants got enough sleep and woke refreshed (never/rarely/sometimes versus often/always); time spent studying a week; house officer or registrar; and in the past fortnight, the number of days worked (
10 or >10); hours worked (no excess—50–70 hours in 1 week and <50 hours in the other, small excess—50–70 hours in both weeks or >70 hours in 1 week and <50 hours in the other, large excess—>70 hours in 1 week and 50 hours in the other); at least two shifts >14 hours in 1 week (yes/no); shifts longer than rostered (none, 1–5 days, >5 days); worked
24 hours straight (yes/no); days on call (none, <3 a week,
3 in at least 1 week); night duty (
1 night a week, 2 in at least 1 week, 3 in at least 1 week); breaks <10 hours (none,
1 a week, >1 in at least 1 week); 24‐hour breaks (
2 a week,
2 in 1 week and 1 in the other, 1 in both weeks, 0 in at least 1 week); roster changes (none, in 1 week, in both weeks); sleep 23:00–07:00 (1 good (
6 nights) week and 1 at least medium (4–5 nights) week, 2 medium weeks, at least 1 bad (
3 nights) week); nights with enough sleep (1 good week and 1 at least medium week, 2 medium weeks, at least 1 bad week); adequate supervision at work (never/rarely/sometimes, often/always); education on coping with shift work (yes/no); and access to a bed at work (yes/no).
Each factor with a significant relationship (p<0.05) to a dependent variable was considered for inclusion in the logistic multiple regression models for that variable. For each outcome measure, two logistic multiple regression models were run. The first model included demographic factors and the work‐related variables. The second model included demographic factors and the total risk score, which captures the particular combination of work‐related factors experienced by each participant. Only participants with complete data for all variables in the model were included (for ESS >10, n
1049; for feeling close to falling asleep at the wheel in the past 12 months, n
1053; for recalling fatigue‐related error(s) in clinical practice in the past 6 months, n