The audited population included all adult outpatients (aged 18 years and older) who visited a defined panel of primary care providers at 10 sites at least once in 1995. In addition to age and sex, the following data were abstracted: billable preventive procedures (Papanicolaou test, screening mammogram, cholesterol test, tetanus immunization, fecal occult blood test, and sigmoidoscopy), total visits to primary care providers, insurance type (HMO, FFS, Medicare, Medicaid, or workers’ compensation), and visit type (acute care–only or scheduled preventive care visit). Cholesterol testing as part of a chemistry panel was counted separately from single cholesterol determinations or lipid profiles, which were classified as screening.
Computer Audit Validation Using Medical Record Reviews
At one study site, the computer audit results were validated by manual review of 200 randomly selected medical records. The validation audit found good agreement (±2.5%) for documentation of preventive procedures; the proportion of positive criteria for the computerized audit was always within 2.5% of that for the manual chart review of the same patient population. Computerized visit frequency counts were highly correlated with medical record review counts (R = 0.93; P <.0001). Medical record review also showed that, except for sigmoidoscopy, approximately 98% of audited preventive procedures had been performed for screening, not for symptoms. (Sixty-seven percent of sigmoidoscopies had been done for screening.)
A separate review of a stratified sample of 245 medical records4 found evidence that screening was performed outside the system of care (and not recorded in the transaction files) in 0.5% or fewer of preventive procedures; the only exception was Papanicolaou tests, which had been obtained outside the system in 13.8% of women aged 20 to 29 years and 4.5% of those aged 30 to 39 years.
Statistical Testing
We explored univariate associations between age, sex, visit frequency, insurance type, and visit type using analysis of variance for continuous dependent variables and the Fisher exact test for binary categorical variables. We used logistic regression to test whether preventive services were independently associated with those same variables. For the statistical testing, visit frequency was defined as all visits to primary care providers in 1995, and presence of a visit scheduled specifically for preventive care was coded as positive if one occurred during 1994 or 1995 (P <.05 was reported as significant).
Results
Thirty-five family practice physicians, 33 general internists, 15 obstetrician/gynecologists, and 12 physician assistants supervised by the primary care physicians made up the primary care provider panel. This panel encountered 75,621 outpatients at least once in 1995. Characteristics of this patient group are presented in Table 2.
Age was positively associated with men (mean = 47 years vs 45 years for women), visit frequency (R = .073), HMO insurance (mean = 45 years vs 41 years for FFS) and acute care–visits (mean = 48 years vs 43 years for preventive care visit) (P <.001 for all comparisons). Women were positively associated with visit frequency (mean = 3.4 visits per year vs 2.6 visits for men) and preventive care visits (52% of all women scheduled such visits vs 22% of men ).
HMO membership was equal for both sexes. Visit frequency was positively associated with HMO membership (mean = 3.2 visits vs 2.6 for FFS) and preventive care visits (mean = 3.1 visits vs 3.0 for acute care only) (P <.001 for both). HMO members scheduled more preventive visits than FFS patients in 1995 (36% and 31%, respectively; P <.001).
Table 3 presents the results for the Papanicolaou test, mammography, cholesterol, tetanus immunization, fecal occult blood testing, and sigmoidoscopy. In general, older patients received more preventive services than younger patients; differences by sex were inconsistent. Patients with 3 or more visits had consistently more positive criteria (7% to 12% more) than patients encountered only once or twice in 1995. HMO patients had 6% to 13% more screening than FFS patients. Associations with visit type were greatest: Patients who scheduled at least one preventive visit had 9% to 45% more positive criteria than patients seen only for acute care. Because of the large sample size, all these differences were statistically significant. Even a trivial positive association of age and mammography use achieved statistical significance (P = .01) in the multivariate model. Logistic regression analyses showed that, for every preventive service audited, the associations of visit frequency and type were statistically independent (P <.001 in all cases).
Table 4 illustrates the profound interaction between insurance type and visit type. HMO patients who scheduled a preventive visit had the highest rates of screening, while FFS patients seen only for acute care had rates that were 14% to 50% lower. FFS patients who scheduled a preventive visit had rates approaching the high rates of screening of the HMO/preventive visit group. Interestingly, HMO patients seen only for acute care had low rates of screening, closer to those of FFS/acute care–only than to the HMO patients who scheduled a preventive visit.