Participants and setting
All children admitted to the ARCADE cohort were eligible for participating in the trial, which was carried out as a substudy of ARCADE. ARCADE participants were recruited from 14 general practices in 2004, in The Netherlands. Admission criteria and baseline characteristics have been published elsewhere . Briefly, children were between 1 and 5 years old had a high risk of developing asthma. Children were identified by an electronic search in computerized records of the general practitioners. 771 children’s parents signed the informed consent and participated in ARCADE. They were followed until the age of 6 years.
Interventions and control
There were three intervention arms and one no-intervention control arm. In the first intervention group (V), ten €12.50-worth gift vouchers were raffled after each questionnaire round among participants who returned a completed questionnaire within 2 weeks of the date of sending. Ten participants in the second intervention group (VD) had a chance to receive a €12.50-worth gift voucher and, additionally, four families could receive a daytrip to the popular Dutch amusement park ‘Efteling’ with the whole family, if they had returned all questionnaires at the end of the study irrespective of the number of reminders. The third intervention group (D) could only receive the above-mentioned daytrip, but no gift voucher. The fourth group (C) was the no incentives control group.
In the daytrip trial, groups D and VD together were classified as intervention group ‘offered a daytrip’, and groups V and C together served as a reference group not offered a daytrip. Similarly, in the gift voucher trial, groups V and VD together were the intervention group ‘offered a gift voucher’ and groups D and C served as a reference group not offered a gift voucher.
Participants were randomized to one of the four groups at baseline for the duration of the follow up. The letters accompanying all questionnaires were for the most part identical for the four groups; except that the letters in the three intervention groups featured an additional paragraph, with a bold printed heading (see additional file for the letter (translated from Dutch into English).
The follow-up of the trial nested in the ARCADE study would last for 2 years, and participants received a maximum of 4 questionnaires, depending of their age at onset. For example, children older than 5 years at onset received only 2 questionnaires until they reached the age of 6 years.
Participants were informed at the start of the study about receiving a questionnaire every 6 months (T0, T6, T12 and T18). Every 12 months (T0 and T12) they received a questionnaire containing 130 multiple-choice questions about quality of life and airway problems . In between (T6 and T18) they received a shorter questionnaire, containing 38 multiple-choice questions about the child’s health-related quality of life only. If a participant did not answer within 2 weeks, they received a postal reminder. Participants received a postal personalized letter and questionnaire consisting of a bright coloured cover and a stamped return envelope. The personalized letter had a logo of the academic hospital and the specific ARCADE study logo, both in red ink and signed by the researcher.
Study design and randomization
The trial had a factorial design (see Figure 1). The four different groups were divided in 2 × 2 groups: Daytrip yes/no (daytrip trial) and gift voucher yes/no (gift voucher trial). An independent member of our staff (JM) assigned participants to one of the four groups according to a computer generated randomization list, with VisualBasic for Application in Microsoft Access. Randomization, stratified by general practice, was performed before sending the first questionnaire. We used a computer random number generator (the seed was the system’s timer) to select 4 random permuted blocks. The block size varied dependent on the number of patients with informed consent per practice location.
Patients were unaware of the trial and the group sizes, but were fully informed about ARCADE. The study was approved by the Dutch Central Committee on Research Involving Human Subjects (CCMO).
The primary endpoint was the effect on overall response rate. For secondary outcomes we analyzed ‘proportion with 100% returned with or without reminder’, ‘probability of non-response’, ‘probability of withdrawal’, ‘percentage of returned questionnaires’ and ‘overall number of reminders sent’.
An additional analysis in the gift voucher trial was done on response rate per questionnaire. Outcome was calculated with repeated measurements logistic regression every 6 months.
All data were collected anonymously in an Access database. The dates of sending the questionnaire or reminder and receiving the questionnaire or reminder were noted.
In total 15% of the values in the dataset (ranging from 6-42%) on covariates (level of education, allergy or asthma of the parents and IgE-serum results) were missing. The missing values were imputed using iterative chained equations (ICE; 20 imputation sets) . These imputed data sets were used to be able to study potential intervention by intervention and intervention by covariate interactions with more power. Using the outcome ‘100% returned’, intervention interactions were assessed where we used alpha = 0.05 to define intervention interaction. We intended to analyse the data as two separate trials provided that there was no positive interaction between the two interventions. No positive interaction was found. Next, within each trial we assessed intervention-‘level of education’ interaction, intervention-ethnicity interaction, and interaction between intervention and the number of questionnaires (related to the child’s age at baseline) and all possible three-way interactions of those covariates to see, for example, if the intervention effects were different for respondents with low education and non-Dutch ethnicity. These interactions too were absent. We calculated risk differences and numbers needed to treat and their 95% confidence intervals without further adjustment for covariates. Proportions were tested using Chi-squared statistics (outcomes ‘100% returned’, 100% returned, no reminder’, ‘non-response’ and ‘withdrawal’), quantile regression of the median for the non-normally distributed variables (outcomes ‘number of reminders’ and ‘% returned’) and multilevel linear regression to take into account the repeated questionnaires every six months within each respondent.
All statistical analyses were performed with Stata 10.1 (Stata Corp., College Station, TX, USA).