Evidence from several studies, including small-scale experimental investigations, suggests that primary care office systems such as flowsheets, reminder systems, and chart "flagging" can increase utilization of several cancer prevention services [7, 8]. Little research has been done, however, on the associations between detailed aspects of a practice's entire system for handling mammography referral and utilization rates. Our main goal in this study was to determine which practice and office system characteristics, if any, predict mammography utilization in a diverse sample of primary care practices.

In the "patient path model" described by Pommerenke and Dietrich [9], four spheres of influence are identified that affect the probability of preventive care being delivered as a patient moves through a medical encounter. These spheres include the patient, the patient's environment, the physician, and the physician's environment. By studying women and physicians participating in a health maintenance organization (HMO) in an area with relatively high levels of education, we reduced the variability in mammography utilization that might be attributable to factors other than those in the physician's practice environment. Our study focused on practice-related factors that intersect at critical points with the patient's path toward regular mammography. These factors, which can come into play before, during, or after an individual encounter, include methods for alerting staff to the need for a mammogram; counseling and education for patients; methods for directly reminding patients of the need for a mammogram; scheduling procedures; and results reporting and follow-up. In addition, we examined potential predictors in two other areas: general practice characteristics, such as patient-to-staff and staff-to-doctor ratios, medical specialty, and percentage of patients receiving Medicaid; and features of the mammography center used most often by the practice.

The study sample consisted of primary care practices participating in an independent-practice model HMO in which subscribers are "assigned" to a single primary care provider. Use of this HMO population eliminated cost as a barrier to mammography because annual screening is fully covered for subscribers older than 50 years and also permitted calculation of accurate utilization estimates for each practice based on computerized claims data.

Methods

Top Methods Results Discussion Author & Article Info References

We conducted structured telephone interviews with nonphysician staff from 132 primary care practices (representing 321 physicians) between July and September 1991. The key respondent was identified by the interviewer as the person most familiar with the organization of mammography services in the practice. In about 75% of the cases, the key respondent was an office manager or nurse manager. Eighty-four percent of the respondents had more than 2 years of employment in the practice. In some cases, multiple respondents were identified to ensure the most accurate completion of the entire interview. To be eligible for interview, the practice had to have at least one physician registered as a primary care provided with Central Massachusetts Health Care, an independent-practice association HMO.

The questionnaire consisted of four parts: 1) demographic data on the practice, including staffing patterns, estimated average number of patients per week, and types of payment; 2) office systems for preventive services, counseling, and patient education; 3) specific procedures for mammography scheduling and referral; and 4) attitudes and knowledge about breast cancer and screening. To enhance the validity and detail of information on the radiology centers where mammography was done, a brief, separate interview of managers at centers used by the practices was also conducted. This interview ascertained the scheduling, results-reporting, and educational procedures of each center. To derive indexes measuring the "patient load" for staff, we calculated ratios of patients seen per week to the number of staff full-time equivalents. A similar index, weekly patients per clinician (physician, nurse practitioner, or physician assistant), provided a measure of patient load for clinicians. "Staff support" available to clinicians was indexed in three ways: as ratios of clinical staff (nurses and medical assistants), nonclinical staff (clerical and administrative), and total staff to the number of clinicians.

To measure mammography utilization, we calculated a utilization index for each practice using billing data from the HMO's management information system. The index is the proportion of women 50 to 74 years old who were continuously assigned to a practice during a recent 12-month period and who received a mammogram during the period. For most women, the 12-month study period ended on 31 January 1991; other women with 12-month periods ending no earlier than 30 June 1990 were also included in the analysis. Data on the age of each woman and her duration of membership in Central Massachusetts Health Care were also obtained.

We used SPS statistical software (SPS Inc., Chicago, Illinois) to analyze the relation of practice variables to the utilization index. In the analysis, utilization for each practice was weighted by the inverse of its variance to account for variance heterogeneity (varying "information value") because of the different number of women assigned to each practice. We conducted bivariate and stratified analyses to initially assess the relation of survey and demographic variables to utilization, using the Student t-test. Finally, we fitted various weighted least-squares linear regression models to assess the independent effects of predictors on utilization.

Results

Top Methods Results Discussion Author & Article Info References

Of 157 eligible practices, 25 were unable or unwilling to complete the interview process, yielding a response rate of 84.1%. Representatives of 26 of 28 (93%) eligible mammography centers were interviewed. Practices on which survey data were collected represent 42% of the obstetrician-gynecologists, 39% of the family physicians, and 31% of the general internists listed in the HMO market area, which includes approximately 600 000 persons. Of the 132 practices surveyed, 111 had at least one eligible woman, encompassing a total of 4378 women (mean age, 56.4 years). The median number of eligible women per practice was 21; 70.3% of the practices had 10 or more eligible women.

The mean mammography utilization for all 111 practices in the analysis was 0.45. Thus, 45% of the women older than 50 years received a mammogram during their most recent 12-month period of continuous assignment to a practice. The relation between categorical predictors and weighted utilization is shown in Table 1. Among procedures used by the practices to facilitate mammography referral, the use of flowsheets and reminders (by mail or telephone) each had strong crude associations with mammography utilization. "Chart flagging" (reviewing and marking charts of patients who are scheduled for office visits) and chart review of patients not necessarily scheduled for visits were not associated with increased utilization. Use by a practice of an increasing number of these tools for alerting staff was not associated with a clear trend toward increasing utilization. Patient education procedures, including prevention counseling by a nurse, were not strongly associated with utilization. Practices with posters and videos on mammography also did not have higher utilization.

The results of a linear regression analysis relating the utilization measure to continuous variable indicators of "patient load" and "staff support" are shown in Table 2. A substantial inverse association with utilization existed for the ratio of patients seen per week to the number of staff, and a smaller but still significant inverse association was found for the ratio of patients seen to the number of clinical providers. The ratios of clinical staff, nonclinical staff, and total office staff to providers were not consistently associated with utilization.

The results from a weighted least-squares linear regression model incorporating utilization as the dependent variable are shown in Table 3. Covariates in the model were chosen based on a priori judgments regarding the relation of various factors to utilization as well as on the magnitude and stability of the coefficients after addition of other covariates to the model. Type of practice (group or solo), use of a particular urban mammography center, and a low percentage of Medicaid patients in the practice had the strongest association with mammography utilization, each increasing predicted utilization by approximately 9% to 12%. Use of flowsheets and the scheduling of mammograms by the patients themselves were each independently associated with 7% to 9% increases in utilization. The mean age of eligible women in the practice was inversely associated with utilization; the proportion of women receiving mammograms was approximately 2% lower for each additional year of mean age. Use of reminders to patients (5% increase), the presence of only internists on staff (5%), and patient-to-staff ratio (1% decrease for each 10 additional patients per staff) had notable effects that approached statistical significance.

Several other models, including potential predictors with significant bivariate relations such as urban or academic practice, gave nearly identical coefficient estimates for the variables in the final model. The combination of related factors into additive scores, for example, for the number of staff reminder procedures used in a practice (flowsheets, chart flags, and chart review) or for the number of patient education modalities available, did not provide a better fit or a more informative model. To describe the flowsheet and reminder effects in more detail, we grouped practices according to whether they used flowsheets only, reminders only, or both and compared utilization in the multivariate model to practices that used neither tool. Practices with flowsheets only had 9.5% higher adjusted utilization (P = 0.03), practices with reminders only had 6.6% higher utilization (P = 0.05), and practices using both tools had 11.4% higher utilization (P = 0.002). Several interaction terms involving staff ratios and specific office tools were tested, and none were found to be significant.

The appropriate R² statistic for the weighted model was 0.51, indicating that the final model accounted for 51% of the variance in utilization. Figure 1 is a scatterplot of the observed versus predicted utilization for each practice, with points grouped according to practice size. Accuracy of prediction is positively related to practice size; thus larger practices tend to group closer around the diagonal than do smaller practices. Among practices with at least 30 eligible women, our model predicted utilization within 5% of actual utilization for 45% of the practices and within 10% for 66% of the practices.

View larger version (26K): [in this window] [in a new window]   Figure 1. Observed and predicted proportion of women receiving a mammographic evaluation by category of practice size. See Table 3 for regression model coefficients.

Discussion

Top Methods Results Discussion Author & Article Info References

Our study demonstrated that several specific characteristics of primary care office systems are related to the proportion of women 50 to 74 years old with HMO coverage who receive mammography at the recommended interval. Although several characteristics showed strong crude associations with utilization, the number of important predictors was reduced through simultaneous adjustment in a linear regression model. The model accounts for 51% of the variation in mammography coverage among primary care practices without considering direct measures of the attitudes of physicians or patients toward mammography screening. Our finding could be strengthened by replication in other geographic areas of the country, where office characteristics might play a larger or smaller role relative to attitudinal factors. Although the HMO that furnished the setting for our study does not collect detailed demographic data on its subscribers, we assume, based on the market area and the high frequency of HMO enrollment through the workplace, that white race, education beyond high school, and current employment were over-represented in our practice sample when compared with the general U.S. population of women.

Our finding of higher mammography utilization in group practices than in solo practices is consistent with previous reports on cancer screening procedures in general [10]. The persistence of this effect after adjustment for several factors reflecting the availability of resources and systems for organizing preventive care could be the result of failure to completely control for these resource differences or the result of behavioral differences between group and solo physicians. The finding of lower utilization among HMO patients in practices that had a higher percentage of Medicaid patients could also be due to resource differences that were not controlled for in the model. The higher competing demands of acute and psychosocial care in these practices can place greater stress on the resources available. Also, to the extent that a high percentage of patients on Medicaid serves as a marker for the socioeconomic status of patients in the practice as a whole, HMO patients enrolled in these practices might also be less informed or less motivated to pursue regular mammography. Increasing mean age of the mammography-eligible women in a practice was negatively associated with utilization, consistent with evidence from previous studies that older women report less frequent use of mammography [11]. The mean age of women in our study was only 56.4 years; other studies with a greater number of older women could find an even larger age effect in the higher age ranges.

The largest independent predictor of mammography utilization was use of a single mammography center (Center 17) located in a private hospital. Center 17 is unusual in several respects: It conducts the most mammography examinations of the 26 centers used by the study practices, provides immediate readings and interpretations to patients from radiologists on site, and maintains a low rate of no-shows (8%) by telephoning patients before a scheduled appointment. Wolosin [12] has shown that sending a reminder card after scheduling an appointment for mammography can substantially improve compliance. Center 17 was also designed, following the advice of physicians and patients, as a facility dedicated to mammography with its own parking, registration, and waiting room. It is staffed by female volunteers and technicians who work only with mammography patients.

We evaluated immediate readings as a separate predictor for utilization and found that its coefficient and effect on the model were similar but somewhat less than those obtained for Center 17 itself as a predictor. It was not feasible, given the collinearity in our sample, to distinguish the extent to which the beneficial effect of the mammography center was due to immediate readings. We chose the specific mammography-center variable for the final model presented, however, because the larger effect seen for the center itself suggested that the other attributes of the center were contributing to its effect. These findings indicate that mammography-center design and policy can increase compliance with screening guidelines, perhaps by creating a positive experience among women who could return for repeated mammograms, by establishing a favorable "word-of- mouth" reputation among women who had never had mammograms, and by developing positive experiences regarding the mammography process among referring physicians. Other attributes of mammography centers, such as distance from the primary care office and waiting time for scheduling a routine mammogram, were not associated with utilization. A policy, at the primary care office, of contacting patients to communicate negative results was not associated with increased utilization.

Among the office tools used to alert staff or patients to the need for scheduling a mammogram, flowsheets on the charts had the largest effect on utilization. Other studies of cancer prevention procedures and mammography in particular have found a beneficial effect for flowsheets of computerized prompts [8, 13, 14]. The modest increase of 6.6% in mammography utilization associated with noncomputerized flowsheets should be evaluated with regard to the relatively low cost and effort involved in using them. In contrast, the smaller benefit associated with sending reminders to women enrolled in the practice involves relatively higher costs for a practice, unless reminder notices could be generated from a central source such as the HMO computer. Practices using both flowsheets and reminders appeared to gain additional utilization over practices using only one of these tools. An additive effect would be expected because flowsheets and reminders involve separate aspects of the "patient path." Flagging the charts of patients scheduled for appointments and reviewing (auditing) a sample of charts for mammography compliance had negligible effects on actual utilization. A score variable indicating the number of staff-alerting tools used by a practice did not perform significantly better than the individual variables in any model.

We had initially hypothesized that having office staff schedule a mammogram ("passive scheduling") would increase utilization when compared with having the patient herself contact the mammography center ("active scheduling"). This would seem particularly likely if, as in our sample, most practices contacting the mammography units did so while the patient waited in the office. The finding of a greater than 8% increase in utilization in practices that had patients do the scheduling was therefore unexpected. We explored several ways in which this finding could have been confounded. For example, practices whose patients had higher income or educational levels might be more likely to allow patients to schedule their own mammograms directly. Active scheduling by patients was associated with a low percentage of Medicaid patients in the practice (P = 0.01); however, the effect of active scheduling remained substantial after inclusion of the Medicaid variable in the model. We also assessed whether practices with active scheduling were more likely to conduct follow-up to determine if patients actually show up for mammograms. We found, however, that the proportion of practices conducting follow-up on no-shows was high (approximately 80%) in both active and passive scheduling practices. It is certainly possible that some confounding due to educational and cultural factors remains even after control for percentage of Medicaid patients, which was the best marker for these factors available to us. Alternatively, it seems plausible to hypothesize that active scheduling, accompanied by follow-up for no-shows, could boost mammography utilization by allowing women to be more engaged in the scheduling process.

Our finding that practices composed entirely of internists had higher mammography coverage than did those composed entirely of family physicians is consistent with small effects of similar nature seen by other investigators [15]. In physician-level models not presented here, consideration of the number of years since graduation from medical school and physician gender produced only a slight reduction in the estimated effect for internists. Consideration of whether family physicians were residency or nonresidency trained could reduce differences further, but we were not able to obtain such data. Patient education offerings, including availability of pamphlets, posters, and videos related to mammography, were not associated with increased utilization in the multivariate models. A score variable for the number of educational offerings to patients was also insignificant. McPhee and colleagues [16] have previously reported that simply providing educational materials to patients about mammography did not influence utilization. Prevention counseling by nurses was the only educational factor that had even a modest effect when added to the model; however, this was not easily distinguished from a null effect in our sample.

The number of patients seen relative to staff size—an index of staff workload—was negatively associated with mammography utilization, as predicted. However, after adjustment for various confounders in our model, the size of this effect was relatively small. An increase of 100 weekly patients per office staff member predicts only a 9% decrease in utilization. Other measures of workload, such as weekly number of patients per physician or other providers, gave similar results. This appears to be consistent with survey results in which physicians report that insufficient time is not a major barrier to mammography screening [3]. Measures of staff support available to providers, including clinical staff such as medical assistants and nurses, did not produce significant results in either the crude or adjusted analyses. Interaction terms, designed to detect whether certain tools such as flowsheets or reminders had different effects depending on the workload of the office staff, did not appear to be important in any of the models.

The observational nature of this and similar studies demands caution in drawing causal interpretations. Despite efforts at adjustment for confounding, it remains possible that the associations contained in our model are biased by failure to control for unmeasured confounders. For example, physicians who feel strongly about the value of mammography and are highly persuasive with their patients might adopt more office tools for promoting mammography, regardless of the effectiveness of such tools. Our data cast doubt on this as a major source of bias, however, because the availability of various patient education materials, which should be correlated with physician zeal, was not associated with utilization. The analysis we have presented relied on measures of the availability rather than the use of office systems because self-reported estimates of use were considered unreliable without confirmatory chart auditing or direct observation of the practice. The effects of actual use of various office systems on mammography are likely to be somewhat greater than the effects we report.

Observational studies complement experimental studies in that they focus on the effectiveness of interventions in actual office environments rather than their short-term efficacy under trial conditions. Our hope is that less expensive studies such as ours will lay the proper groundwork for large-scale, randomized trials of office system interventions to boost mammography rates. These trials could use routinely collected, computerized outcome data covering defined populations to enhance their accuracy and cost efficiency.