Health Care Systems for Tracking Colorectal Cancer Screening Tests
4. Assessment of the Intervention (continued)
PRISM Outcome Domain
LVHN study staff recruited 25 primary care practices (20 intervention and 5 control), with a total of 100 providers (81 intervention and 19 control) to participate in the SATIS-PHI/CRC intervention. Subsequent to recruitment and allocation to study arm, but prior to the start of the intervention (during the time period of waiting for OMB clearance), 5 practices assigned to the intervention arm, with a total of 17 providers, dropped out. Thus, we had only 15 intervention practices with 64 providers to participate in SATIS-PHI/CRC; none of the control practices dropped out.
To recruit these practices, LVHN study staff needed to contact 43 practices, for a successful recruitment rate of 20/43 or 46.5 percent.xv Table 4.3 presents the number and distribution of the 25 intervention, dropout, and control practices and their providers, by affiliation, specialty, size, and location of the practice. These distributions by attribute for intervention and control practices are generally comparable without inclusion of the dropout practices.
To assess the representativeness of the practices participating in the SATIS-PHI/CRC intervention relative to the pool of practices in the LVPHO pool from which we recruited, we compared the distribution of practices in the pool and in the intervention. Table 4.4 presents this comparison for practices classified by affiliation (the only practice attribute for which we had comparison data). We recruited and included all three of the LVHN clinics in the pool to ensure inclusion of the urban core and a sufficient number of Medicaid and uninsured patients. LVPG practices are also overrepresented, and MATLV and independent practices are underrepresented, largely because LVPG practices were more willing to participate than were the others.
Based on information provided by LVHN study staff, we know that several practices approached during the recruitment process were reluctant to participate given the stringent economic environment and current transitions to EMR systems. They feared that participation would require too many resources, whether financial, information technology, or staff time. This fear was especially prominent among practices not owned or operated by LVHN. Similarly, the practices that initially agreed to participate but later dropped out said that they were too busy or too short staffed and they would need to be compensated for the time they spent participating. One dropout practice was sold to a competing network after agreeing to participate and thus no longer met the eligibility criteria for participation.
Overall, our intervention had high program fidelity, as we were able to implement most intervention elements as planned. We successfully recruited all 26 practices (1 for the pilot, 20 for the intervention, and 5 for the control). As previously noted, five intervention practices dropped out after recruitment but prior to the start of the intervention. We also were able to successfully recruit a stool test kit supplier for the intervention.
We were successful in conducting academic detailing in all of our intervention practices. We found it easier to schedule academic detailing sessions in some of the practices than others, but all were scheduled and completed as planned. We also had planned to conduct academic detailing sessions in the control practices at the end of the intervention period. Due to time and budget limitations, we were not able to roll out the intervention to all control practices. But the central entity still plans to distribute the intervention toolkit and other intervention materials to the control practices at the end of the study.
As noted, we needed to conduct an academic detailing booster to further emphasize guideline-recommended screening modalities. We disseminated this booster successfully. In addition, based on lessons learned from the pilot, we disseminated a packet of information to office managers explaining their role and what they needed to do, including what to do with the stool test kits. However, we learned that the office manager did not always distribute this information to others in the practice, so there was some lingering confusion.
We also were successful in conducting our electronic record reviews, although our experience varied by practice and by data source. The central entity had a more difficult time conducting the electronic record reviews in some practices than others. This problem caused us to send the patient mailings in several waves, depending on the patient mailing lists that were available.
In addition, the central entity could not retrieve complete lists of all age- and screening-eligible patients from each practice. Some practices could only provide a list of PHO-insured patients. One practice could not link its patient lists with its PHO insurance data, so it could not tell who was a PHO patient and who was not. Therefore, the final list of study participants was not a complete list of all age- and screening-eligible patients in these practices; it was a subset based on available data. As we have noted, the organization was not accustomed to population-based public health interventions of this magnitude, which affected their ability to conduct the electronic record reviews.
We were able to mail all intervention materials to eligible patients, and we were able to mail stool test kits to all patients who requested one. However, there were times when patients responded to the SEA indicating they did not want to participate but the subsequent mailing was already on its way. We also experienced some delays in sending out the patient mailings, due to a greater than originally anticipated labor requirement to generate and disseminate the mailings. Finally, we found that the SEA was successful in identifying patients who were not eligible for the intervention.
For tracking patients and their screening results, we successfully implemented the intervention, but these elements of the intervention did not always produce the planned outcomes. For example, we provided a Screening Tracking Sheet to the office manager of each practice. However, we know that none of the practices used this tool to track their patients. Some practices used their own methods for tracking, but there were no practices that were able to share these methods with us. We were able to conduct nearly all the chart audits as planned. In a few cases, charts were not available for auditing.
For our final intervention step, providing feedback to practices, we successfully sent all Feedback Forms to the practices as intended. The stool test lab also sent all results directly to each practice. We reviewed the records and found that all patients with positive stool tests were notified of their results and followup was recommended or performed. However, the central entity performed these reviews manually, as there was no way to systematically search the electronic records for this information. For negative screening results, we could not determine if all patients were notified by their practice, becuase this information was not readily available in the data sources.
In terms of ongoing maintenance of the intervention, based on our informal conversation with LVHN, their organization's leadership is interested in adopting this intervention as part of its future patient-centered medical home efforts. Their leadership was originally focused on chronic disease registries, but they are now looking at screening, especially CRC. It is likely that this intervention will become part of the organization's wider medical home efforts.
Based on the postintervention practice focus groups, some of the practice respondents said the intervention had not changed the way in which they practiced. Others said that it had increased their awareness or convinced them to use the FIT and other new screening modalities. One respondent said that the intervention had stimulated more tracking of individual patients who began the screening process. Many of the focus group participants said they did not know the true impact of the intervention yet. However, most said they would participate in the intervention again if given the opportunity.
While we have minimal data in terms of patient maintenance, we did receive a response from the patient focus groups that indicated the patient was interested in continuing the intervention next year. The patient wanted to be able to receive another stool test kit request card without having to go to the doctor. In the future, one way to accurately assess patient maintenance would be to see if patients who screened by stool test this year continue to rescreen annually.
A total of 10,627 patients participated in the SATIS-PHI/CRC intervention, 7,965 in either of the two intervention arms and 2,662 in the control arm (Figure 3.1 and Table 4.2). An initial pool of 12,808 patients was identified as potentially eligible based on an initial review of available electronic records. We deemed a total of 10,982 patients to be eligible for participation. From that group, we excluded 355 patients based on their opting out of the intervention or our inability to deliver intervention material to them by mail.
The 10,627 participants are 96.8 percent of deemed eligible patients and 83.0 percent of all initial potentially eligible patients. The initial pool of 12,808 patients may not be fully representative of the target population of all potentially eligible patients within LVPHO primary care practices. The practices participating in the intervention are not fully representative of all LVPHO primary care practices and several participating practices were not able to provide all of the electronic data required for identifying all of their potentially eligible patients (Table 3.1).
The flow diagram in Figure 3.2 illustrates the reasons for eliminating potentially eligible patients from the initial pool. Table 4.5 presents a more detailed look at eliminated patients. We eliminated some patients because we deemed them to be ineligible based on subsequent review of electronic records, charts audits, or responses to the SEA form. Others were eligible but excluded because they opted out or had undeliverable addresses (and thus could not participate in a mailed intervention).
We eliminated equivalent proportions of intervention and control group patients as ineligible based on electronic records and chart audits. We eliminated additional intervention group patients as ineligible based on responses to the SEA form, a source not available to control group patients to whom we did not mail SEA forms. Further, we excluded otherwise eligible intervention patients (but not control patients) based on sources not available to control patients. These additional sources of elimination may have caused the intervention and control groups to lose comparability.
Table 4.6 addresses the representativeness of intervention group patients who participated in SATIS-PHI/CRC by comparing their distribution on various attributes to intervention group patients whom we deemed to be ineligible or whom we excluded. Compared to participating patients, excluded patients were more likely to be female, older, and Medicare insured whereas deemed ineligible patients were largely comparable. All three groups were generally comparable on practice characteristics.
Finally, Table 4.7 addresses representativeness by comparing the practice affiliation of all LVPHO patients ages 50-79 to intervention and control group patients included in the SATIS-PHI/CRC study. These distributions are not comparable for two reasons. First, we purposively sampled practices to ensure inclusion and a balanced proportion of each practice affiliation type in intervention and control arms of the study. Second, practices varied in their ability to provide electronic records data on their patients. In particular, patients from hospital clinics were overrepresented in the study. Especially in the control arm, patients of LVPG practices were also overrepresented in the study, and MATLV and other independent practices were underrepresented.
We assessed the effectiveness of the SATIS-PHI/CRC intervention by studying its effect on the screening rate for CRC, followup rate for positive screens, and provider and practice behavior.
Table 4.8 presents results comparing intervention group and control group CRC screening rates for stool test, colonoscopy, and any modality. It also includes results comparing screening rates for patients randomized to receive the stool test kit directly and those randomized to receive a mail-back card to request a kit. The table presents results separately for:
- All participating practices.
- Only the two intervention practices in which patients were randomized.
- Only those practices with the most complete or moderately complete electronic data.
- Only those practices with EMR systems used for electronic record review.
We conducted the latter two analyses with only practices having more complete electronic data and EMR systems in order to exclude practices in which detecting colonoscopy screening was more problematic due to incomplete or less detailed data. We expected this data deficiency to affect our ability to detect colonoscopies, because we could only detect them through electronic records or chart audits. We did not expect it to affect our ability to detect stool test screens, because the participating clinical lab reported FIT results directly to us.
Results for all practices and for practices with more complete electronic data include a comparison of the intervention group to the control group using an "adjusted" control group denominator. We adjusted these denominators to account for lack of information from the SEA form, undeliverable mailings, and opt-outs that we had for intervention patients. The adjustment consisted of reducing the denominator (number of eligible participants) of control practices by a number proportionate to the number of SEA-ineligible, undeliverable, and opt-out patients in intervention practices. We then used this reduced denominator for purposes of this analysis only.
Results for the two practices with randomized patients only include a comparison of the kit intervention group to the card intervention group because neither is a control group practice. Results for practices with more complete electronic data and with EMR systems only include a comparison of the card intervention to the control because these practices do not include those with patients randomized to receive the kit intervention. We adjusted all odds ratio (OR) confidence intervals (CIs) and p values in Table 4.8 for possible clustering effects of assigning an entire practice's patients to intervention or control study groups. For all practices, both card intervention patients and kit intervention patients had significantly higher odds of having a stool test and of having any screening test compared to controls. This result persisted even after adjusting the control group denominator.
Although card intervention patients had somewhat higher odds than controls to be screened by colonoscopy, this result was not statistically significant. Our ability to detect colonoscopies in the practices with poorer data affected this result, as revealed in the results of the analyses restricted to practices with better data. For these better data practices, the ORs all improve in favor of the intervention group relative to the "all practices" ORs and in several instances are significant at the p<0.10 level.
Kit intervention patients, who all came from practices with poorer data for colonoscopy, had significantly lower odds of being screened by colonoscopy than did controls but this result is almost certainly a result of the poorer data. An alternative explanation could be that directly providing the test kit to patients increases the odds that stool test rather than colonoscopy will be the screening modality used.
In general, these OR results comparing intervention patients to controls are indicative of the intervention's effectiveness. The Table 4.8 ORs for any screening test are all statistically significant and compare favorably with the OR reported in the study on which we based the screening intervention portion of SATIS-PHI/CRC (Myers, et al., 2007). A comparison of the card intervention and the kit intervention with controls yields overall ORs for all practices for any screening test of 2.31 (CI=1.39-3.85) and 2.16 (CI=1.40-3.35), respectively.
The OR reported by the comparison study for the standard interventionxvi was 1.68 (CI=1.25-2.53). Our ORs decrease somewhat when we adjust the control group denominator (1.89 and 1.77 for the card and kit interventions, respectively) but remain significant and comparable to the OR reported by Myers, et al. (2007). Further, our ORs for practices with better data for detecting colonoscopies are somewhat higher than for all practices.
These effectiveness ORs presented in Table 4.8 would likely have been even higher if we had been able to send stool test kits directly to all intervention patients rather than having to use a card intervention for most of them. Within the two practices with randomized patients, those receiving the kit directly were 3 times more likely to be screened by stool test (OR=3.16; CI=2.40-4.16) compared to those having to mail back a card to request a kit. Kit intervention patients were also 2 times more likely to be screened by any test (OR=2.53; CI=2.14-3.00). This result strongly suggests that had we used the kit intervention exclusively, the intervention would have been more effective overall.
Regardless of the OR results, the actual screening rates observed were substantially lower than we expected and substantially lower than those reported by Myers, et al. (2007). According to Table 4.8, screening rates by any test were 8.6 percent for card intervention patients and 8.1 percent for kit intervention patients (with a combined rate of 8.5 percent) and 4.7 percent for control patients using the adjusted denominator. By comparison, the Myers, et al., study reported screening rates of 46 percent for the standard intervention group and 33 percent for controls.
There are several potential reasons for our lower observed screening rates. Our measurement of screenings was not as robust as in the comparison study. We know that we have both numerator (detecting screenings) and denominator (detecting and eliminating ineligible subjects) errors involving both intervention and control patients. The original study was able to more closely review existing medical record and administrative data (and to conduct telephone interviews with patients) to identify truly eligible patients and to case find screenings. Further, to qualify for inclusion in that study, patients had to not only be eligible for screening but also had to complete a baseline survey and thus to have already indicated a willingness to participate.
Another reason is that our intervention and observation period was much shorter than in the original study. We implemented a single round of the SATIS-PHI/CRC intervention and observed subsequent screenings over an 8-month period. The original study implemented two successive rounds of intervention and observed subsequent screenings over a 24-month period.
A third potential reason is the nature of the study setting. The original study was based in a medical school-run teaching clinic that is very different than community practices. It is likely that offering screening to patients is taught as a recommended behavior in the medical school teaching clinic. The original study's control group screening rate was 33 percent compared with under 5 percent for the Lehigh Valley practices.
Even though our observed screening rates were substantially lower than those reported by the Myers, et al., study, the percentage increase, as reflected in the ORs, we observed exceeded that of the earlier study. The intervention group in the Myers, et al., study screened at a rate about 40 percent higher than controls (46 percent compared to 33 percent). Our intervention group screened at a rate 81 percent higher than controls (8.5 percent compared to 4.7 percent). In general, we conclude that our implementation of the SATIS-PHI/CRC intervention in Lehigh Valley primary care practices achieved screening rate outcomes that were comparable with those of the study used as the basis of our intervention.
We also calculated age-sex standardized screening rates to adjust for differences in age-sex distributions between intervention and control groups. Table 4.9 displays the results. Even after controlling for age and sex through standardization, the screening rates for both intervention groups significantly exceed those of the control group for screening by stool test and any test. The card intervention group colonoscopy screening rate also exceeds that of the control group. As was true for the unstandardized results in Table 4.8, the age-sex standardized colonoscopy screening rates for the kit intervention group is significantly lower than that of the control group for the same likely reasons as cited above for the Table 4.8 OR results.
We investigated the effect of returning a completed SEA form on screening. Since only intervention patients received an SEA form, we could not include control patients in this analysis. Table 4.10 shows that the 1,131 intervention patients who returned a completed form (out of the 7,965 study-eligible patients to whom we sent one by mail) were significantly more likely to be screened by stool test (OR=5.90, CI=4.73-7.36), colonoscopy (OR=5.60, CI=4.48-7.01), or any test (OR=6.66, CI=5.63-7.87).
Returning an SEA form likely indicates an interest in or willingness to participate in a screening program. This result also suggests that capturing patient interest through a motivating introductory letter and getting them involved with the intervention effort early in the process with the simple SEA form could increase screening rates. Note that the screening rate for any screening test for patients responding to the SEA form (27.59 percent) more closely approximates the rate reported by the Myers, et al., study that required patients to have completed a baseline survey in order to qualify for inclusion.
We also investigated the variation in screening rates by practice. Table 4.11 presents the results. For this analysis, we excluded from the study the two control practices and five intervention practices that each had fewer than 60 patients because of the statistical instability of rates with small denominators. We also excluded the three intervention practices that had poor colonoscopy data from the analysis of colonoscopy screening rates. We separately calculated rates for control practices using unadjusted and adjusted denominators.
To help maintain comparability between intervention practice stool test screening rates, we kept the rates for card intervention patients separate from those for kit intervention patients in the two practices with patients randomized between interventions. We only compared the card rates in these practices to those in the other intervention practices. To assess variation within intervention practices and within control practices, we calculated group means and standard deviations for intervention practices and for control practices for each screening modality. We then computed a coefficient of variation (CV, equal to the standard deviation divided by the mean) for them. To assess variation between intervention and control practices, we calculated a comparison of means for them using an F test to estimate the statistical significance of any differences.
There is generally substantial variation between practices, with CVs ranging from 0.23 to over 1.4. The intervention group means are all considerably higher than the control group means for each modality and the difference between means in each case is statistically significant. On average, intervention group rates are significantly higher than control group rates.
The next set of five tables (Tables 4.12-4.16) present bivariate and multivariate logistic regression analyses of CRC screening by stool test, colonoscopy, and any test. The first three tables present results for analyses of patients exposed to the SATIS-PHI/CRC intervention to indicate what types of patient and practice characteristics are associated with a greater or lesser likelihood of being screened by a given modality. Table 4.12 presents the bivariate results for six patient characteristics and five practice characteristics. For three of the patient characteristics (marital status, perceived health status, and education), we only had data for a limited subset of patients, so we excluded them from the multivariate analyses. All of the clinic practices were urban, making it inadvisable to include both practice affiliation and location in multivariate models. We decided to include location and exclude affiliation.
Table 4.13 presents the full multivariate model for each screening modality, incorporating the three patient characteristics for which we had data for more than just a subset of patients and the four remaining practice characteristics after we eliminated affiliation. Table 4.14 presents a reduced model incorporating only those characteristics found to be statistically significant at the p<0.10 level for a given modality in the full multivariate model. We included the data source characteristic for practices in the reduced model regardless of whether it was significant in the full model for a given modality because we wanted to be certain to control for data quality. Both of the multivariate tables present Hosmer-Lemeshow Χ2 goodness of fit tests along with odds ratios for the screening tests.
The next two tables present similar multivariate analyses but also include study arm as a practice characteristic. These analyses allowed us to assess the effectiveness of the intervention by examining the ORs by each intervention arm relative to the control arm. As with the intervention practice analyses only, Table 4.15 presents the full multivariate model and Table 4.16 presents the reduced model. However, since we did not have insurance coverage data for control patients, we could not include that characteristic in these models. As with the previous multivariate model tables, these tables present Hosmer-Lemeshow Χ2 goodness of fit tests along with odds ratios.
The bivariate results presented in Table 4.12 suggest that for stool tests, older patients and those insured through Medicare are more likely to be screened than younger patients or those insured through other health plans. Single patients are less likely to screen than married patients, and patients attending hospital clinic practices are less likely to screen than those attending independent practices.
For colonoscopy, the results show that women are somewhat more likely to be screened than men; Medicaid and uninsured/self-pay patients are less likely to screen than those with commercial insurance; single patients are less likely to screen than married patients; patients attending general internal medicine practices are more likely to screen than those attending family medicine practices; patients attending large practices are more likely to screen than those attending small practices, and patients attending suburban practices are somewhat more likely to screen than those attending urban practices. In addition, as expected, the completeness of colonoscopy data affects the likelihood of detecting screening by this modality.
Patients more likely to be screened by any modality include women, older patients, those who perceive their health to be excellent/very good or good, and those attending general internal medicine practices, large practices, and suburban practices. Patients less likely to be screened by any modality include uninsured/self-pay patients, single patients, and patients attending hospital clinic practices. As with the results for colonoscopy, data completeness affects the likelihood of detecting screening by any means.
The multivariate results in Tables 4.13 and 4.14 reveal that gender is not significant for stool test screening but is for colonoscopy. Conversely, age is significant for stool testing but not for colonoscopy. Medicaid/self-pay patients are less likely to be screened by stool test. Patients of larger practices are less likely to be screened by any modality whereas those of general internal medicine practices are more likely to be screened by colonoscopy. Practice location is significant for colonoscopy, with patients of nonurban practices being more likely to screen. Neither the full or reduced model for stool tests is a good fit (by the Hosmer-Lemeshow test) whereas those for colonoscopy and any test are good fits.
Tables 4.15 and 4.16 reveal that even after statistically controlling for patient and practice characteristics, receiving the intervention significantly increases the odds of being screened for CRC. Both the full and reduced models were a good fit based on the Hosmer-Lemeshow test, with the reduced model being a better fit for stool test screening and screening by any test and the full model being a better fit for colonoscopy screening.
In particular, both the card and the kit intervention have significantly large ORs relative to controls for stool testing in both the full and reduced models, and the card but not the kit intervention also has significantly large ORs relative to controls for colonoscopy in both models. The kit intervention had no significant impact on colonoscopy screening in either model, although those receiving the kit intervention did have a nonsignificant 1.5 times higher odds of colonoscopy screening relative to controls in the reduced model. The inability of the kit intervention to increase the odds of colonoscopy screening is most likely the result of the poor quality of the colonoscopy detection data available in the two practices whose patients received this intervention. Both interventions had significant effects on being screened by any test in both models.
Based on all of the above results for the effectiveness of SATIS-PHI/CRC to increase the odds of being screened for colorectal cancer, and in particular on the results of Tables 4.15 and 4.16, we conclude that the intervention was effective.
Followup of Positive Screens
In addition to increasing the odds of being screened for CRC, an intended effect of SATIS-PHI/CRC was to increase the likelihood that positive screens would be followed up with actions recommended by current screening guidelines. In particular for SATIS-PHI/CRC, which highlighted stool testing and colonoscopy, we expected that positive stool tests would be followed up with a complete diagnostic examination (CDE) using colonoscopy.
During the 8-month SATIS-PHI/CRC study observation period, we observed evidence of 363 stool test screens among intervention and control patients, 348 in the intervention group and 15 in the control group. Of those stool tests, 7 of the 348 (2.0 percent) intervention group screens had positive (abnormal) results and 18 (5.2 percent) had unknown results; the remainder (92.8 percent) were known to be negative (normal). Of the 15 control group stool tests, one (6.7 percent) was positive and none had unknown results; the remaining 14 (93.3 percent) were negative. Thus, during the observation period, we detected only 8 known abnormal stool tests and 18 with unknown results. These small numbers preclude a detailed analysis of effectiveness and especially preclude a comparison between intervention and control group experience.
We did, however, track this small number of positive and unknown stool tests to determine their outcome (without evaluating the comparative followup rate for intervention and control patients). Of the single control patient with a positive stool test, a chart audit revealed that the provider recommended a CDE colonoscopy but the patient refused it. Of the seven intervention patients with known positive stool tests, six (85.7 percent) received a CDE. We could not determine if the seventh patient received a recommendation for a CDE.
We also determined that 2 of the 18 intervention patients with unknown stool test results received CDE colonoscopies. Thus, eight intervention patients received a CDE. Of those eight patients, half had an abnormal finding. This finding does mean that these four patients were diagnosed with CRC but rather that an abnormality of some kind was detected. Of the remaining four CDEs, one was negative; we could not determine the results of the other three.
Provider Knowledge and Behavior
A third intended outcome of SATIS-PHI/CRC was to influence the CRC screening knowledge and behavior of providers receiving the academic detailing component of the intervention so that they were more consistent with current screening guidelines. To test the effectiveness of the intervention to achieve this outcome, we compared the responses of providers to the preintervention and postintervention survey of intervention practices. We focused in particular on responses to the first two sections of the survey.
Section A of the survey asked providers to indicate which CRC screening tests they frequently recommended to their screening-eligible patients and which tests they believed were effective. Section B asked providers to indicate their likely followup actions to a positive stool test and an abnormal flexible sigmoidoscopy finding. Tables 4.17 and 4.18 present the results of this analysis. We did not include comparisons with the survey of the control practices because the number of clinician respondents (N = 5) was too small for analysis.
Table 4.17 indicates that a smaller percentage of postintervention respondents said they recommend fecal occult blood test (FOBT) screening compared to preintervention respondents. A larger percentage of postintervention respondents said they recommend fecal immunochemical test (FIT) screening.xvii This finding suggests that the academic detailing effort to increase FIT testing over FOBT was successful. There were no other statistically significant pre-post differences in response to the question on recommended tests.
We expected to see a decrease in the percentage of clinicians recommending digital rectal examination (DRE), which is not a guideline-recommended screening test for CRC, but that did not occur. On that criterion, the academic detailing was not effective. We did observe an increase in the percentage of respondents indicating that they believed a wide selection of tests are effective for CRC screening.
Although only two screening modalities had significant increases (FIT and stool DNA testing), all others with the exception of DRE increased somewhat, even if not significantly. However, given the small number of provider respondents to each of these surveys (54 to the preintervention survey and 41 to the postintervention survey), the surveys lack the statistical power to detect smaller differences. We also observed a small decrease in the percentage of respondents indicating that they believed DRE to be effective. This result is in the expected direction but is too small to be significant.
Table 4.18 indicates that a higher percentage of postintervention provider respondents than preintervention provider respondents said they would follow up a positive stool test with a recommended action. This finding supports the effectiveness of the academic detailing. But there was no significant change in the percentage of respondents saying they would follow up an abnormal flexible sigmoidoscopy with a recommended action, although the observed difference was in the expected direction.
xv LVHN study staff actually contacted 44 practices and recruited 26 of them: 25 for allocation to a study arm and one to be the site of the pilot intervention.
xvi The Myers, et al. (2007) study implemented a standard intervention and two more tailored interventions. They found that the more tailored interventions did not achieve higher screening rates than the standard intervention alone. We thus based SATIS-PHI/CRC on the standard intervention.
xvii Recall that the survey of practices was anonymous and that we have no way of knowing whether or to what extent the same or different respondents participated in the preintervention and postintervention administrations of the survey. We surveyed the same practices and distributed the survey to the same people within them for both administrations, but we cannot determine who responded to both or to one but not the other. Even though there may be sufficient background demographic information in survey results to allow us to match some pre and post respondents, our IRB protocol would preclude doing so.