Overall, the results indicated that a small cash incentive and appropriate recruitment procedures can produce strong response rates to a lengthy mailed survey among cancer patients. Our overall adjusted response rate was 67%. The rates we report here are slightly higher, but comparable to some prior studies, which have achieved response rates of 60-65% [45, 46]. However, many of the studies that have achieved the highest response rates have used convenience samples or other non-representative populations. In addition, this study was undertaken more than 10 years after the referenced studies, a period when a decline in response rates to all forms of surveys has been a major concern of researchers. The achieved rate is substantially higher than other studies have found for mailed surveys in the last decade [13, 37, 47]. In a review of 141 academic papers describing 175 separate studies published in management and behavioral sciences in the years 1975, 1985 and 1995, an average response rate of 55.6% was estimated . Evidence that a mailed survey drawing from a statewide registry in 2006 was able to achieve this rate despite the reported declines in response rates is noteworthy. The only parallel evidence we could find for mailed cancer patient surveys comes from a study in the Netherlands in 2005 using a single hospital's registry, but it also achieved a response rate of 62% .
Interestingly, the amount of incentive ($3 versus $5) was not a significant factor predicting response. In fact, the response rates are virtually the same between participants who received $3 and participants who received $5. The finding that the high response rate was received with a relatively small incentive amount is encouraging for future research.
While there was a tendency for the shorter questionnaire to earn a higher response rate, this was not a statistically significant difference. This lack of significant difference was surprising, given the drastic difference between the two surveys. The post hoc results suggest that length and incentive amount did not affect item response either.
The adjusted response rates between breast cancer participants and prostate cancer participants are comparable (73% vs. 69.5%, respectively). However, colon cancer participants responded at a lower rate (57%) than both breast cancer and prostate cancer participants. The pattern among the three cancers is consistent with the large mixed mode ACS survey of survivors, for which 42% of breast, 35% of prostate and 30% of colon cancer patients responded . This pattern of response may be partly explained by the differential level of morbidity associated with these cancers, which may affect ability to respond.
An alternative interpretation relies not on the actual inability to respond but the emotions evoked by the surveys. According to the Leverage-Saliency Theory of Survey Participation, the achieved influence of a particular feature is a function of how important it is to the potential respondent, whether its influence is positive or negative, and how salient it becomes to the sample person during the presentation of the survey request . A survey which reminds a patient that he or she has a cancer with relatively higher morbidity and less positive prognosis (colon cancer) may result in a lower response rate than a survey about a cancer evoking emotions about a health condition with a better prognosis (prostate or breast).
Similarly both of these explanations (actual inability and psychological reluctance) may also explain why patients at more advanced stages of disease were less likely to respond (45% for metastatic versus 66% for others). Colon cancer patients were more likely to be at higher stage at diagnosis (14% metastatic versus 5% for breast and 2% for prostate).
The results by demographics provided some valuable information for the sampling plan in the larger subsequent study. For example, because non-whites and those with stage 4 cancers consistently responded less often across conditions, those two groups were over-sampled in order to ensure sufficient sub-group numbers for the larger study. The lack of evidence for interactions between demographics and condition were encouraging. Even if $3 and short surveys were used, the heterogeneity of the sample would not be compromised.
The high rates of response despite secular declines in response rates may have several explanations. Two possible explanations are discussed here: First, the procedures followed those recommended by Dillman quite closely. Second, these questionnaires were sent to cancer patients who had been diagnosed in the previous calendar year. This was a high salience issue for the patients. Also, the questions that were asked were primarily about their experience in trying to choose treatments and survive with their cancers, and about their use of public information sources, as well as medical sources of information. Respondents may have been appreciative of the opportunity to discuss these topics.
These results are highly generalizable to other cancer patient research contexts given the selection of a representative sample and mailed survey implementation followed standard recommended procedures, but subject to the limitations described below.