Exposed members were recruited from sampling of lists of participants in existing
research projects. Projects were identified from a list of studies approved by the Harvard Pilgrim Health Care Human Studies Committee. This list was more comprehensive than that used for the pilot study. A total of 96 studies were identified that were current at the time o f sampling. These were narrowed to those that included patient contact. Sixteen investigators were contacted, several of whom had multiple ongoing studies. A number had studies involving only a few patients, or were at a phase when patient contact by the study team was considered unsuitable by the investigators, either because it may interfere with the study design, or it was too early in terms of patient involvement. Two studies of substance abuse were considered, but were not included because of specific confidentiality clauses in the consent forms. Permission was sought from the principal investigators of the selected studies prior to sampling. Investigators provided lists of participants including either member identification numbers, or name and demographic details. Member identification numbers were used to search the AMRS.
Table 4. Studies included in member survey, research sample
Study Study design, and sampling
method
n
Comparison study of the cost effectiveness of Lovastatin versus stepped-care in the treatment of primary hypercholesterolemia
Multi center, prospective, open label, parallel study. Sample provided by investigators
23
Decision model for hormone replacement therapy use
Personal interview o f peri- menopausal women. All
participants (who were contacted and did not decline)
21
Developing and validating a dietary screening tool for primary care
Questionnaire in person, and blood sample
Consecutive patients
40
Hypertension and automated linkage of outcomes study
Monitoring of compliance with treatment.
Random sample
40
Oncology/hematology chemotherapy and radiotherapy trials
Patients enrolled in a number of treatment protocols. Random sample of list provided by investigators
40
HIV studies Patients enrolled in trials of therapy
for patients with AIDS. Patients contacted prior to survey by investigators
28
Attention deficit hyperactivity disorder study
Study of family, genetic and psychosocial risk factors for girls aged 6-17 years
All Harvard Pilgrim patients involved in study
14
Childhood asthma management program Trial of treatment and side effects in pediatric asthma patients. Random sample
57
Clinical quality management (CQM) pediatric asthma survey
Annual mail survey of pediatric asthma patients, addressing quality of care. Random sample
35
Access to and use of oral rehydration solution
Randomized controlled trial of prescribed oral rehydration solution in pediatric patients. Random sample
40
A control group of unexposed members was identified from the AMRS. Unexposed members were frequency matched with exposed members on the basis of age (pediatric with pediatric, deciles for adults), diagnosis that made the patient eligible for the study (if
relevant), health center, and gender. All controls had been members for at least one year. Criteria for matching are shown in Table 5.
Table 5. Criteria o f matching fo r controls fo r research surveys (in addition to age, gender and health center)
Study Criteria for matching controls
Comparison study of the cost effectiveness of Lovastatin versus stepped-care in the treatment of primary hypercholesterolemia
Diagnosis code for hypertension
Decision model for hormone replacement therapy use
No disease match
Developing and validating a dietary screening tool for primary care
No disease match
Hypertension and automated linkage of outcomes study
Diagnosis code for hypertension
Oncology/hematology chemotherapy and radiotherapy trials
Oncology patients
HIV studies Diagnosis code for HIV
Attention deficit hyperactivity disorder study
Diagnosis code for allergic conditions in pediatric patients
Childhood asthma management program Diagnosis code for asthma in pediatric patients
CQM pediatric asthma survey Diagnosis code for asthma in pediatric patients
Access to and use of oral rehydration solution
No disease match
A number of exposed members were eliminated because they were not Health Centers Division members and could not be matched with a control on a sample site. One case was lost for this reason from each of the following studies: decision model for HRT use, dietary screening tool, HIV studies, oncology studies, oral rehydration, CQM asthma survey, and the Childhood asthma management program.
In total, 188 questionnaires were sent to exposed adult members, and 188 to the sample o f control adult members. The pediatric version of the survey was sent to parents or guardians of 143 exposed pediatric members, and to a sample of 143 parents or guardians of control pediatric members. The total number of questionnaires sent was 376 adult and 286 pediatric.
A follow up letter and second copy of the questionnaire was sent after two weeks to members who had not responded to the first questionnaire. A total o f 238 reminders were sent to adults and 247 to parents or guardians of children. The follow up letter requested return within two weeks, and, as with the first, asked members who were not interested in participating to return the blank questionnaire.
The number o f surveys sent was based on sample size calculations and estimated response rates of approximately 60%. The original sample size calculations suggested that the adult sample should be in the order of 130 exposed patients and 130 matched unexposed patients.
This sample size was estimated using the following principles. The calculations were based on detecting a difference in the responses of two groups to a dichotomous question. The probability of a Type I (a) error was set at 0.05 (one-tailed), i.e. there would be a 5% chance of wrongly concluding there is a difference between the two groups. The Type II (P) error was set at 0.2 i.e. there would be an 80% chance of finding a significant
difference between the two groups.
There was no data from previous studies on which to base estimation of effect size. Discussions with the study advisors and a statistician resulted in basing the sample size calculations on an estimation o f differing response in a proportion of 0.15 to 0.2 of respondents.
Using these parameters, sample size tables were consulted (Hulley and Cummings, 1988). If PI is the proportion of subjects expected to have the outcome in one group and P2 in the second group, then a sample of 130 in each group would be adequate to detect a difference between the two groups of 0.15. This sample size of 260 would also be large enough to determine whether a different response to a dichotomous question by a proportion of 0.2 of respondents is significant (95% confidence interval 0.15, 0.25).
In the pediatric group a sample of 92 exposed and 92 unexposed respondents would provide a sample of 184, large enough to determine whether a different response to a
dichotomous question by a proportion of 0.15 of respondents is significant (95% confidence interval (0.08, 0.23). Detection of a difference between exposed and unexposed groups o f 0.2 would be possible with a power o f 95%.