Margaret J. Gunter, PhD

To compare colorectal cancer (CRC) screening use, including changes over time and demographic characteristics associated with screening receipt, between 2 insured primary care populations.

Clinical and administrative records from 2 large health systems, one in New Mexico and the other in Michigan, were used to determine use of CRC screening tests between 2004 and 2008 among patients aged 51 to 74 years.

 Generalized estimating equations were used to evaluate trends in CRC screening use over time and the association of demographic and other factors with screening receipt.

 Rates of CRC screening use ranged from 48.1% at the New Mexico site to 68.7% at the Michigan site, with colonoscopy being the most frequently used modality. Fecal occult blood test was used inconsistently by substantial proportions of patients who did not meet the definition of screening users. Screening use was positively

and significantly associated with older age, male sex, and more periodic health examinations and other types of primary care visits; at the Michigan site, it was also associated with African American race, married status, and higher annual estimated household income.

Among insured primary care patients, CRC screening use falls short. Further research is needed to determine what factors are barriers to routine fecal occult blood test or colonoscopy use among insured patients who have access to and regularly use primary care and how those barriers can be eliminated.

Commonalities in colorectal cancer screening use were evident across 2 insured primary care populations.

Between one-third and one-half of patients at each site were not up-to-date with screening.

At both sites, younger patients, female patients, and patients with fewer periodic health examinations and other types of primary care visits were at greater risk of being unscreened for colorectal cancer.

Among recommended testing modalities, colonoscopy was by far the most used test among patients who were up-to-date with screening.

 Although fecal occult blood test use was common, adherence to annual testing was a challenge at both sites.

Colorectal cancer (CRC) screening was first recommended by professional organizations for average-risk adults 50 years and older more than 2 decades ago.

Most recently, the US Preventive Services Task Force

 issued evidence-based CRC screening guidelines recommending the use of the following: (1) fecal occult blood test (FOBT) or fecal immunochemical test annually, (2) FOBT or fecal immunochemical test every 3 years combined with flexible sigmoidoscopy every 5 years, or (3) colonoscopy every 10 years. Despite these guidelines and the availability of several effective screening tests, CRC screening remains underused.

Population-based surveys indicate that CRC screening rates have increased steadily over time,

 with the most recent estimates showing that screening rates exceed 60%.

 Less recently, claims-based and medical record—based studies among insured populations point to similar trends but tend to place estimates of the screened population somewhat below those obtained by self-report

 and indicate an increase in the use of colonoscopy relative to other modalities.

Using automated claims and electronic medical record data for a 5-year period ending December 31, 2008, we compared CRC screening use between insured primary care populations receiving care from 2 large integrated delivery systems, one located in southeast Michigan and the other located in central New Mexico. We report annual screening use by modality between 2004 and 2008 and examine demographic and other patient factors associated with screening receipt.

Eligible patient populations were selected from those receiving primary care at 2 health systems. The Michigan health system includes a 900-member multispecialty salaried medical group that staffs 27 primary care clinics in Detroit and surrounding communities. The New Mexico health system is the largest physician-owned multispecialty physician group in the state, with 230 salaried physicians and 70 midlevel care providers. It staffs 10 primary care clinics in Albuquerque and surrounding communities. Both sites are participating in a Centers for Disease Control and Prevention—sponsored randomized controlled trial evaluating an intervention to increase CRC screening among primary care patients. These 2 health systems were chosen for trial participation because they have a large number of eligible patients and

physicians, a high proportion of minority patients, and strong research capacity and data capture capabilities.

Patients eligible for study inclusion were those aged 51 to 74 years on December 31, 2008, and continuously enrolled in a system-affiliated health maintenance organization during the 5-year study period ending December 31, 2008 (ie, the 5-year period preceding intervention implementation). To our knowledge, no randomized controlled trials to increase CRC screening took place at either health system during the study period. All patients with health plan enrollment were eligible regardless of whether their coverage was sponsored by employer, Medicare, or other. Although Michigan patients had copayments for office visits, they had no cost sharing specific to CRC screening tests. New Mexico patients may have had a deductible, minimal copayment, or both for endoscopy procedures depending on their personal coverage and situation. Eligible study patients also had at least 1 primary care visit during the study period (2004-2008). Exclusion criteria included evidence of a history of colorectal carcinoma, ulcerative colitis, Crohn’s disease, or bowel resection. The study was approved by the institutional review board at each participating site.

Automated health system records were used to obtain patient age, sex, race (Michigan site only) or ethnicity (New Mexico site only), residential street address, marital status (Michigan site only), primary care visit frequency, diagnoses, and receipt of CRC screening tests. At the Michigan site, race is recorded as 1 ethnicity was determined using GUESS (Generally Useful Ethnicity Search System), which identifies ethnicity based on surname. GUESS has been shown to be 90% accurate for identifying individuals of Hispanic ethnicity in New Mexico.

Residential street address was combined with zip code—level data for 2009 from the Current Population Survey to construct an annual estimated household income for patients.

International Classification of Diseases, Ninth Revision, Clinical Modification

 diagnostic codes during the 5-year period were used to construct an adapted Charlson Comorbidity Index.

 codes were used to categorize primary care visits during the 5-year period as periodic health examinations (PHEs) or as other types of primary care visits.

We calculated annual testing rates for FOBT, colonoscopy, flexible sigmoidoscopy, double-contrast barium enema, combination of FOBT and flexible sigmoidoscopy, and other modality combinations for 2004 through 2008. Five-year CRC screening receipt was based on the 2002 US Preventive Services Task Force guidelines.

 Patients who received at least 3 FOBTs (

10 months apart) or 1 colonoscopy, flexible sigmoidoscopy, or double-contrast barium enema during the 5-year period were considered screening users. Patients with evidence of colonoscopy receipt between 1999 and 2003 were also considered screening users.

Characteristics of patients receiving care from the 2 health systems were compared using X

 test for differences in age, and Wilcoxon rank sum test for differences in annual estimated household income and primary care visit frequency. For the analysis of time effects, year was represented by 4 indicator variables, and an overall 4-

 test for year effect was performed. For analyses of withinsystem effects, generalized estimating equation logistic regression models were used to account for the nesting of patients receiving care from the same primary care provider. For differences in screening use by patient characteristics, generalized

estimating equation logistic regression models were used for each characteristic by itself and after adjustment for patient age, sex, race/ethnicity, annual estimated household income, frequency of PHEs and other types of primary care visits, and Charlson Comorbidity Index. The Michigan health system models were also adjusted for marital status.

In Michigan, 13.5% of the cohort were older than 65 years, and 54.1% were female (

). Approximately one-third of patients (37.1%) were African American. Most patients (73.4%) were married, and the mean annual estimated house-communities of residence was $76,846. Sixty-three percent of patients had at least 1 PHE during the study period, and 37.0% had at least 2 PHEs in the 5-year period. Compared with Michigan patients, New Mexico patients were significantly older (P <.01) and were more likely to be female, reside in areas with lower annual estimated household income, and have fewer primary care visits. hold income in their communities of residence was $89,809. Most patients (63.4%) had at least 1 PHE during the study period (2004-2008), with 44.0% having at least 2 such visits.

In New Mexico, 46.8% of patients were older than 65 years, and 57.0% were female, while approximately one-third of patients (36.2%) were of Hispanic ethnicity (Table 1). The mean annual estimated household income in their  communities of residence was $76,846. Sixty-three percent of patients had at least 1 PHE during the study period, and 37.0% had at least 2 PHEs in the 5-year period. Compared with Michigan patients, New Mexico patients were significantly older (

 <.01) and were more likely to be female, reside in areas with lower annual estimated household income, and have fewer primary care visits.

, annual CRC screening use (overall and for specific modalities) varied by year at both sites (

 <.01 for all). At the Michigan site, the overall annual CRC screening rate increased steadily between 2004 and 2008. Annual use of colonoscopy increased between 2004 and 2007, and annual use of FOBT increased between 2005 and 2008. At the New Mexico site, the overall annual CRC screening rate increased between 2004 and 2005 and declined thereafter. Annual colonoscopy and FOBT use among New Mexico patients paralleled this overall pattern. At both health systems, only the use of colonoscopy alone or FOBT alone contributed substantively to the observed annual use of screening.

Sixty-nine percent of the Michigan cohort and 48.1% of the New Mexico cohort met the criteria for screening users (

). Most users were screened during the 5-year study period, but 9.3% of Michigan patients and 8.7% of New Mexico patients were classified as screening users based on evidence of colonoscopy receipt during the 5 years preceding the study period. Among patients who met the criteria for screening users, colonoscopy was the most frequently used screening modality; almost 90% of users at each health system met the definition of screening users because of their receipt of a colonoscopy. Regardless of health system, the use of each of the other recommended screening modalities was rare, including repeated use of FOBT (

3) alone or in combination with flexible sigmoidoscopy. Among patients who did not meet the criteria for screening users in the 5-year study period, 22.5% of Michigan patients and 43.7% of New Mexico patients had 1 or 2 (but not 

Patient Factors Associated With CRC Screening Use

At the Michigan site, CRC screening users were significantly more likely to be male (46.9% vs 43.6%), African American (37.4% vs 36.5%), and married (74.9% vs 70.0%)

). Compared with nonusers, screening users were significantly older; 38.7% of users were older than 60 years compared with 29.4% of nonusers. Users were also more likely to reside in communities with higher median annual estimated household income ($90,489 vs $88,318), receive more frequent PHEs and other types of primary care visits, and have more comorbidities as measured by the Charlson Comorbidity Index (1.1 vs 0.9). A similar pattern of significant unadjusted differences was found between CRC screening users and nonusers at the New Mexico site, with 2 exceptions: in New Mexico, there were no significant differences by sex, and patients of Hispanic ethnicity were significantly less likely to have been screened (34.5% vs 37.9%).

Among screening nonusers, approximately half (49.1% at the Michigan site and 55.6% at the New Mexico site) had at least 1 PHE during the study period, and approximately onethird had at least 2 PHEs. Almost 12% at the Michigan site and 13.6% at the New Mexico site had 3 or more PHEs during the 5-year period yet remained unscreened. Furthermore, among patients who did not meet the definition of screening users, the mean numbers of other types of primary care visits during the 5-year period were 8.5 (Michigan) and 9.1 (New Mexico) (data not shown). Among patients who completed 1 or 2 FOBTs but not 3 or more (and thus did not meet the criteria for screening users), the mean numbers of other types of primary care visits during the study period were 9.8 at the Michigan site and 9.6 at the New Mexico site (data not shown).

When patient factors were simultaneously controlled, screening use was positively and significantly associated with older age, male sex, and more PHEs and other types of primary care visits at both sites (

). At the Michigan site, African American race, married status, and higher annual estimated household income were also positively associated with screening use. Patients aged 56 to 65 years at the Michigan site were approximately 50% more likely to have been screened for CRC compared with those aged 51 to 55 years, and patients aged 66 to 74 years were more than 60% more likely to have been screened compared with those aged 51 to 55 years. At the New Mexico site, the difference was most pronounced for those aged 71 to 74 years, who were almost 90% more likely to have been screened for CRC compared with those aged 51 to 55 years. Increases in PHE frequency were also associated with dramatic increases in the likelihood of screening. At the Michigan site, patients who had 3 PHEs

during the 5-year period were 4.5 times more likely to have been screened compared with patients who did not have a PHE. At the New Mexico site, patients who had 3 PHEs during the 5-year period were 2.6 times more likely to have been screened compared with patients who did not have a PHE.

Although we found that overall screening trends differed at the 2 health systems studied, with an increase in screening over time at the Michigan site and an initial increase and then a decrease at the New Mexico site, we found remarkable similarities in modality use and in patient factors associated with CRC screening use. Among both populations, most CRC screening users were recipients of colonoscopy. Yet in any given year during the study period, sizable proportions of patients used FOBT screening at both sites. Likewise, although we found a strong positive association between primary care visit frequency (particularly PHEs) and CRC screening use, we also found that substantial proportions of patients who did not meet the criteria for screening users had 2 or more PHEs over the 5-year period. These findings point to considerable

opportunities to improve CRC screening rates among insured primary care populations.

Our findings that almost 90% of CRC screening users met the definition of screening users by having received a colonoscopy are consistent with results of other medical record and claims reviews.

Such evidence is consistent with data from a national survey of physicians demonstrating that almost all physicians recommend colonoscopy to average-risk patients, in many cases as the only screening option.

including research conducted at the Michigan site studied herein, showed that CRC screening discussions in primary care tend to focus on colonoscopy. However, such practice is inconsistent with studies that point to variation

among patient preferences for CRC screening modalities. For example, a study by Hawley and colleagues

 found that 37% of patients prefer colonoscopy screening, 31% prefer FOBT, and 9% prefer sigmoidoscopy. In addition, patient preferences may vary depending on the patient subpopulation.

 Therefore, it is unclear whether our finding that colonoscopy represents the dominant screening modality is a result of patient choice and is consistent with patient preferences.

Frequent use of FOBT was consistent across the 2 sites; substantial proportions of patients (between one-fourth and one-third) completed an FOBT each year. However, most patients utilizing FOBT did not use it consistently. Therefore, few patients met the criteria for screening users by having at least 3 FOBTs within the 5-year period. Among patients considered nonusers of CRC screening, more than 22% at the Michigan site and more than 43% at the New Mexico site used FOBT once or twice. Previous research demonstrated that annual adherence to FOBT is a common challenge.

We found that patients who used FOBT once or twice (but not 

3 times) during the 5-year study period had a mean of almost 10 primary care visits (or 2 per year). Therefore, development of primary care clinic—based processes to ensure annual use

of testing among FOBT users, including promotion of opportunistic screening, may improve CRC screening use among insured primary care populations.

 we found that the receipt of more primary care visits was associated with a greater likelihood of screening use. Nonetheless, more than one-fourth of patients who did not meet the definition of screening users had 2 or more PHEs. Although we cannot know for certain whether CRC screening was recommended by a physician during these visits, data from an observational study

between 2007 and 2009 at the same Michigan health system as that studied herein showed that 94% of patients due for CRC screening who had a PHE received a physician recommendation for CRC screening. Such results, along with findings by others,

 illustrate the necessary but insufficient role of physician recommendation in the CRC screening process and the importance of improvement in the effectiveness of provider recommendations for evidence-based care.

There are several important limitations to this study. First, we relied on automated data, which precludes our discerning whether tests were ordered for screening or diagnostic purposes or whether a colonoscopy was ordered because of another positive test result. This could overestimate screening use and confound observed differences in screening among populations if certain subgroups are more prone to the use of diagnostic vs screening tests. Second, we were limited to a 5-year observation period of continuous enrollment. Although colonoscopies performed in the 5-year period before our observation period were considered when determining screening status, patients who were not continuously enrolled during that time may have received colonoscopy screening elsewhere. Third, although the patient populations studied were large and diverse, they were all insured; while their care was delivered by 2 health systems only, we were unable to systematically evaluate institution-level factors that may have influenced screening use. Fourth, care should be taken when generalizing findings to uninsured patients, those not visiting

a primary care provider, or populations otherwise different from those studied herein. In particular, the high median annual estimated household incomes among the communities in which study patients resided should be noted. However, the national Healthcare Effectiveness Data and Information Set (HEDIS) rate for CRC screening use in commercial health plans during the same period was estimated to be about midway between our 2 observed rates (ie, 58.7%).

Articles

Colorectal cancer screening use was similar in 2 divergent primary care populations. Colonoscopy was the most frequently used modality; FOBT was used inconsistently.

Colorectal Cancer Screening Use Among Insured Primary Care Patients

To describe the proportion of patients receiving drugs with a narrow therapeutic range who lacked serum drug concentration monitoring during a 1-year period of therapy and to identify patient characteristics associated with lack of monitoring.

Ambulatory patients (n = 17 748) at 10 health maintenance organizations who were receiving ongoing continuous drug therapy with digoxin, carbamazepine, divalproex sodium, lithium carbonate, lithium citrate, phenobarbital sodium, phenytoin, phenytoin sodium, primidone, quinidine gluconate, quinidine sulfate, procainamide hydrochloride, theophylline, theophylline sodium glycinate, tacrolimus, or cyclosporine for at least 12 months between January 1, 1999, and June 30, 2001, were identified. Serum drug concentration monitoring was assessed from administrative data and from medical record data.

Fifty percent or more of patients receiving digoxin, theophylline, procainamide, quinidine, or primidone were not monitored, and 25% to 50% of patients receiving divalproex, carbamazepine, phenobarbital, phenytoin, or tacrolimus were not monitored. Younger age was associated with lack of monitoring for patients prescribed digoxin (adjusted odds ratio, 1.86; 95% confidence interval, 1.39-2.48) and theophylline (adjusted odds ratio, 1.58; 95% confidence interval, 1.23-2.04), while older age was associated with lack of monitoring for patients prescribed carbamazepine (adjusted odds ratio, 0.59; 95% confidence interval, 0.44-0.80) and divalproex (adjusted odds ratio, 0.50; 95% confidence interval, 0.38-0.66). Patients with fewer outpatient visits were also less likely to be monitored 

A substantial proportion of ambulatory patients receiving drugs with narrow intervals between doses resulting in beneficial and adverse effects did not have serum drug concentration monitoring during 1 year of use. Clinical implications of this finding need to be evaluated.

The goal of therapeutic drug monitoring is to guide dosing by means of drug concentration measurements. Therapeutic drug monitoring is useful for drugs that lack correlation between dose and pharmacodynamic properties, for drugs that have nonlinear correlations between dose and effect, and for drugs that have a narrow therapeutic range (NTR) between the dose necessary to achieve beneficial effects and the dose that causes serious adverse effects when there is a direct concentration-effect relationship. Individualizing the drug dosage and the dosing interval can minimize the toxicity and maximize the therapeutic benefit of NTR drugs.

The first step toward individualizing therapy is to evaluate the drug concentration in the body, often from a serum sample. Although there is controversy about whether monitoring should be routine, as well as about the frequency and timing of therapeutic drug monitoring to maximize beneficial drug effects,

there is general agreement that therapeutic drug monitoring is useful in avoiding drug toxicity. For example, drug concentration monitoring is recommended as a quality-of-care indicator for patients taking phenytoin, phenobarbital, divalproex sodium, and carbamazepine.

 Concentration monitoring of many NTR drugs is routinely available at healthcare laboratories, but little is known about the frequency of monitoring among ambulatory patients.

We undertook a study to assess drug concentration monitoring in ambulatory patients receiving NTR drugs. The objectives of this study were to describe the proportion of patients dispensed NTR drugs who did not have drug concentration monitoring during a 1-year therapy period and to identify patient characteristics associated with lack of monitoring. The results of this study provide information to guide thoughtful establishment of quality-of-care indicators related to NTR drug monitoring in ambulatory patients.

This retrospective cohort study of ongoing NTR drug therapy was designed to assess rates and correlates of serum drug concentration monitoring among ambulatory members of 10 health maintenance organizations in geographically diverse US locations. We studied monitoring practices for the following NTR drugs: carbamazepine, cyclosporine, digoxin, lithium, phenytoin, phenobarbital, primidone, procainamide hydrochloride, quinidine, tacrolimus, theophylline, and divalproex (valproic acid). The first study objective was to assess the proportion of ambulatory patients receiving continuous ongoing therapy with an NTR drug who did not have at least 1 drug serum concentration monitored within a 1-year period. The second study objective was to evaluate possible associations between lack of NTR drug monitoring and patient age, sex, outpatient visits, chronic diseases, diagnoses, selected socioeconomic status variables, and hospitalizations. Finally, the accuracy of using automated data in identifying drug concentration monitoring was assessed.

The participating organizations comprise the HMO Research Network Center for Education and Research in Therapeutics, which has been described elsewhere.

 In brief, the center includes staff, group, network, independent practice association, and mixed-model health maintenance organizations that serve racially and ethnically diverse populations and in 2000 provided healthcare for approximately 7 million people in more than 1000 locations. The institutional review board of each participating organization approved this study.

The study sample was drawn from a data set of 2 020 037 individuals, consisting of approximately 200 000 randomly selected health plan members from each of the 10 organizations. The sampling scheme and demographic distribution of this population have been previously described.

We identified patients who received an NTR drug of interest between January 1, 1999, and May 31, 2000, and who had continuous health plan membership with pharmacy benefits during the study period, disregarding gaps of less than 60 days. The study sample was limited to prevalent users who had continued drug dispensings of and ongoing therapy with an NTR drug. To avoid inadvertent inclusion of new drug users in the study cohort, prevalent ongoing therapy was defined as beginning with the second dispensing of that drug in the data set and continuing from the second dispensing date for 12 months or longer. Continued dispensings were present when no interval between prescription refills was greater than the dispensed days' supply plus 1.5 times the dispensed days' supply. A dispensing gap was ignored if it was less than 1.5 times the dispensed days' supply. For example, a patient dispensed a 30-day supply met the criterion of continued dispensings if no more than 75 days (30 days + [1.5 Ãƒâ€” 30 days]) elapsed between the date of one dispensing and the date of the subsequent dispensing. NTR drug dispensings were identified using National Drug Code numbers. For all eligible patients, we used automated health plan data to ascertain dates of health plan membership and drug dispensings.

Identification and Categorization of Drug

Lack of laboratory monitoring was defined as failure to perform drug concentration monitoring within the 1-year period of ongoing therapy, allowing up to a 1-month grace period (ie, monitoring within 13 months was considered as monitoring within 1 year). To accommodate patients entering the cohort between January 1, 1999, and May 31, 2000, monitoring data were collected for January 1, 1999, through June 30, 2001. Whether the laboratory test was performed was assessed from the presence of an administrative claim for the test. The laboratory test date was the date the test was performed or the result was reported. Laboratory test codes were defined at each site according to that site's method of coding the test (eg, 

 codes). One site was unable to contribute drug concentration monitoring data for carbamazepine, digoxin, cyclosporine, and phenytoin.

Medical records were randomly selected for review to assess the accuracy of administrative data. Because of resource limitations, this review included medical records of patients dispensed cyclosporine, digoxin, and carbamazepine. Medical record abstraction was considered the gold standard, and abstraction data were compared with administrative data to determine the sensitivity, specificity, positive predictive value, and negative predictive value of administrative information about drug concentration monitoring.

The health maintenance organization automated databases were used to identify the patient age (on the date of the first NTR drug dispensing), sex, outpatient visits, chronic diseases, diagnoses, and hospitalizations. The presence of specific chronic diseases was determined using the Chronic Disease Score method by Clark et al.

To evaluate characteristics that predicted lack of monitoring, hospitalizations and outpatient visits that occurred within 6 months before the study period were identified. Outpatient visits include clinic appointments, visits for laboratory testing only, and emergency department visits. Geocoding was used to provide surrogate patient-level measures of socioeconomic status. Residential street address was combined with census-block level data from the 2000 US Census data to construct proxies of patient race, education, and poverty.

The number of unique patient-drug therapy combinations and the number of patients were tabulated. Descriptive statistics were computed to characterize patients, drug dispensings, and drug concentration monitoring for each drug cohort. The proportions of patients taking each NTR drug who received drug concentration monitoring and those who did not receive drug concentration monitoring were tabulated overall and by health plan, age group, sex, outpatient visits, hospitalizations, Chronic Disease Score, race, education, and poverty categories. Statistical significance of differences was tested using the Wilcoxon rank sum test or the χ

The association between patient characteristics and drug concentration monitoring was evaluated using generalized estimating equation logistic regression modeling, with site as a cluster variable.

The initial generalized estimating equation model for each drug included the following variables with 

< .05 in the univariate analysis: age group, sex, outpatient visits (in increments of 5 visits), hospitalizations, the presence or absence of selected diagnoses, and the 3 linear socioeconomic status variables. The Chronic Disease Score was not included in any of the initial models because it was correlated with outpatient visits. We used the backward selection method for all generalized estimating equation models. Customary residual and effect statistics were examined to assess model fit and to evaluate outliers. Analyses were performed using SAS version 8.2 or 9.1 (SAS Institute Inc, Cary, NC).

Almost 18 000 patients (n = 17 748) met the eligibility criteria for the study; these patients represented 18 821 individual patient-drug therapy combinations (5.7% of patients received > 1 study drug). Patient characteristics are given in 

. Patients prescribed digoxin (n = 7153), divalproex (n = 2209), carbamazepine (n = 1876), phenytoin (n = 2232), theophylline (n = 2195), and lithium (n = 1487) accounted for 91% of the study patients. The median number of drug dispensings during 1 year ranged from 6 for digoxin to 12 for cyclosporine (data not shown).

The proportions of patients who were not monitored were low (14%) for cyclosporine and lithium (

). However, for patients prescribed digoxin, theophylline, procainamide, quinidine, or primidone, 50% or more were not monitored. Between 25% and 50% of patients prescribed divalproex, carbamazepine, phenobarbital, phenytoin, or tacrolimus did not have a serum concentration evaluated within the 1-year period.

Nine participating sites contributed data for all study drugs. At 8 of these sites, 36% to 48% of patients lacked monitoring; at the 9th site, 62% of patients were not monitored. Because the 10th site did not contribute data for 4 NTR drugs, the percentage of patients without monitoring for all NTR study drugs could not be determined.

A consistent association between patients' median age and lack of monitoring was not observed in the univariate analysis. Patients prescribed digoxin (77 vs 74 years, 

= .02), or theophylline (65 vs 61 years, 

< .001) who were less likely to be monitored were younger, while patients prescribed carbamazepine (45 vs 49 years, 

< .001) who were less likely to be monitored were older. For all study drugs, patients who were less likely to be monitored had a lower median number of outpatient visits. This difference was significant for digoxin (10 vs 8 visits, 

< .001), and phenobarbital (5 vs 4 visits, 

< .001) difference (but unlikely a clinically significant difference) in the median percentages of monitored patients who had at least a high school education (89% [5th-95th percentiles, 64%-98%]) compared with patients who had at least a high school education who lacked monitoring (88% [5th-95th percentiles, 62%-98%]). Because patient race was correlated with poverty and with education (Spearman rank correlation coefficient [r] range for race and poverty for individual drugs, -0.43 to -0.53; and r range for race and education for individual drugs, 0.36-0.45) and because poverty and education were correlated (r range for poverty and education for individual drugs, -0.31 to -0.63), race and poverty were not considered further in the analysis.

Logistic regression analysis was used to evaluate associations between patient characteristics and lack of monitoring for the 6 prescribed drugs that accounted for 91% of the study patients (

). When other characteristics were adjusted for, the patient characteristics most consistently associated with monitoring status continued to be age and outpatient visits, while the numbers of chronic diseases and recent hospitalizations were not associated with lack of monitoring for any of these drugs. Younger patients prescribed digoxin were more likely than the oldest patients prescribed digoxin to lack monitoring. Other factors associated with monitoring status of digoxin were female sex, the number of outpatient visits, and a diagnosis of arrhythmia or heart failure. These factors were associated with a lower likelihood of lacking monitoring (ie, patients with these factors were more likely to be monitored).

For patients prescribed theophylline, those in the youngest age group were more likely to lack monitoring than those in the oldest age group (Table 2). Fewer outpatient visits and a diagnosis of chronic obstructive pulmonary disease were associated with lacking theophylline monitoring.

In contrast to the relationship between younger patient age and the presence of monitoring that was observed for digoxin and theophylline, patients prescribed carbamazepine or divalproex who were in the oldest age group were more likely to lack monitoring (Table 2). For carbamazepine and divalproex, a seizure diagnosis or a mental health diagnosis was associated with a lower likelihood of lacking monitoring, as was an increasing number of outpatient visits.

Because patients prescribed carbamazepine or divalproex who did not have a seizure diagnosis or a mental health diagnosis had characteristics that were different from those of individuals with these diagnoses (

), we analyzed these patient groups separately. For carbamazepine, the association between being in the oldest age group and having a lower likelihood of monitoring was significant in patients with a mental health diagnosis (odds ratio, 0.24; 95% confidence interval, 0.11-0.54 for the youngest age group). Similarly, for divalproex, the association between being in the oldest age group and having a lower likelihood of monitoring was significant in patients with a mental health diagnosis (odds ratio, 0.47; 95% confidence interval, 0.32-0.69 for the youngest age group; and odds ratio, 0.53; 95% confidence interval, 0.36-0.77 for the second youngest age group), as well as for the youngest age group of patients who had neither a seizure diagnosis nor a mental health diagnosis (odds ratio, 0.42; 95% confidence interval, 0.22-0.83 for the youngest age group).

The sensitivity of claims data compared with medical record data for drug concentration monitoring was excellent (89% [47/53] for digoxin, 93% [70/75] for carbamazepine, and 95% [63/66] for cyclosporine). The specificity varied (94% [34/36] for digoxin, 78% [14/18] for carbamazepine, and 50% [11/22] for cyclosporine). The positive predictive value, the percentage of patients who had serum drug concentration monitoring according to administrative data that was also documented in medical records, was high (96% [47/49] for digoxin, 81% [60/74] for carbamazepine, and 85% [63/74] for cyclosporine). The negative predictive value, the percentage of patients who did not have serum drug concentration monitoring according to administrative data that was also not documented in medical records, was good (85% [34/40] for digoxin, 74% [14/19] for carbamazepine, and 79% [11/14] for cyclosporine).

Surprisingly high proportions of patients prescribed digoxin, theophylline, phenytoin, carbamazepine, procainamide, quinidine, primidone, divalproex, and phenobarbital (all medications with an NTR) do not have drug concentrations monitored at least yearly. In contrast, lithium and cyclosporine drug serum concentrations were monitored in most patients. Monitoring is recommended for these drugs because life-threatening toxicity can result when concentrations are elevated, therapeutic efficacy can be lost when concentrations are low, and correlations between drug dosages administered and concentrations achieved can be poor.

The most consistent patient characteristic associated with lack of monitoring was a low number of outpatient visits, a finding that is intuitive. For example, it is feasible that the high monitoring rate observed with cyclosporine is related to the fact that patients prescribed this drug have a high number of outpatient visits (median, 11 visits in a 6-month period) (Table 1).

Age was also often associated with lack of monitoring, but the association between age and monitoring varied by drug, with younger age being a predictor of lack of monitoring for patients prescribed digoxin and theophylline, and with older age being a predictor of lack of monitoring for patients prescribed carbamazepine and divalproex. The absence of a specific seizure diagnosis or mental health diagnosis for patients prescribed carbamazepine or divalproex increased the risk for lack of monitoring, as did the absence of an arrhythmia or heart failure diagnosis for patients prescribed digoxin and the absence of a chronic obstructive pulmonary disease diagnosis for patients prescribed theophylline.

Dosages of digoxin used to manage arrhythmias are often higher than dosages used to manage heart failure. Therefore, one could argue that monitoring is more important for patients with arrhythmias because their risk of drug toxicity is greater. Yet, we found that the presence of either diagnosis was similarly protective against lack of monitoring (Table 2).

Patients prescribed carbamazepine who have seizures or a diagnosis such as bipolar disorder are at higher risk of drug toxicity because they are treated with higher drug dosages than patients prescribed carbamazepine for an indication such as trigeminal neuralgia.

It is feasible that clinicians are less concerned about efficacy and toxicity monitoring in patients prescribed carbamazepine for diagnoses other than seizures or bipolar disorders. The same rationale can be applied to patients prescribed divalproex who do not have a seizure diagnosis or a mental health diagnosis. Unfortunately, this study was not designed to evaluate whether a relationship existed between dosage and monitoring, nor was it designed to differentiate between drug concentration monitoring that was not performed because it was not ordered vs because the patient did not obtain the ordered test.

Our work did not evaluate the appropriateness of drug concentration measurements relative to the need, timing, and interpretation of monitoring.

We acknowledge the importance of these factors in evaluating quality of care, but we believe that the information herein is meaningful even without information about the appropriateness of monitoring. Our findings document a widespread absence of drug concentration monitoring. Monitoring drug concentrations is viewed as a quality measure associated with avoiding preventable drug-related morbidity and disease exacerbations by organizations, including the National Committee for Quality Assurance in its Health Employer Information Data Set.

The findings of our study can be used by organizations to improve rates of drug concentration monitoring through implementing guidelines that target the patient risk groups we identify herein as lacking monitoring.

We found good sensitivity, specificity, positive predictive value, and negative predictive value of administrative data compared with medical record data across the organizations that participated in this study. However, the generalizability of our results is limited by the fact that quality of care and the accuracy of administrative data vary across organizations. The quality of diagnostic and procedure coding depends on institutional experience with and emphasis on coding, as well as contractual arrangements between health plans and institutions.

There is a dearth of information about whether drug concentration monitoring reduces adverse outcomes in patients. The evidence we present herein of low rates of drug concentration monitoring should prompt investigations of clinical and economic outcomes in ambulatory patients who are prescribed NTR drugs and who do not receive drug concentration monitoring. Patient outcomes should be used to guide the refinement of existing NTR drug monitoring quality-of-care indicators in ambulatory patients.

We thank Parker Pettus, MS, for leading the data management and computer programming required for this study. We thank Kimberly Lane, MPH, Andrea R. Paolino, MA, and Deanna Merrill, PharmD, MBA, for project management and medical record abstraction coordination. We acknowledge the programmers from each of the health maintenance organizations. We also acknowledge the helpful input provided by Laurie Crounse, MPH, and David McClure, MS, on the analytic plan for this work. Finally, we thank Paul H. Barrett, MD, MSPH, for his thoughtful review of a draft of the manuscript.

From the Clinical Research Unit (MAR, NMC) and Pharmacy Department (EAC), Kaiser Permanente of Colorado, and University of Colorado School of Pharmacy (MAR, EAC), Denver; Meyers Primary Care Institute, Fallon Foundation, and University of Massachusetts Medical School, Worcester (SEA), and Harvard Pilgrim Health Care and Harvard Medical School (SRS, RP) and Channing Laboratory, Brigham and Women's Hospital and Harvard Medical School (KAC, RP), Boston; Henry Ford Health System, Detroit, Mich (JEL); Kaiser Permanente Northwest Center for Health Research and Oregon Health Sciences University, Portland (AF); Lovelace Clinic Foundation, Albuquerque, NM (MJG); HealthPartners Research Foundation, Minneapolis, Minn (WWN); and Group Health Cooperative Center for Health Studies and Department of Epidemiology, University of Washington, Seattle (RLD).

This study was supported by cooperative agreement U18 HS 11843 and the HMO Research Network Center for Education and Research in Therapeutics, funded by the Agency for Healthcare Research and Quality, Rockville, Md.

The contents of this article are solely the opinions of the authors and do not necessarily represent the official views of the Agency for Healthcare Research and Quality, which did not participate in the design or conduct of the study; the collection, management, analysis, or interpretation of the data; or the preparation, review, or approval of the manuscript. Dr Raebel had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Address correspondence to: Marsha A. Raebel, PharmD, Clinical Research Unit, Kaiser Permanente of Colorado, PO Box 378066, Denver, CO 80237-8066. E-mail: marsha.a.raebel@kp.org.

Monitoring of Drugs With a Narrow Therapeutic Range in Ambulatory Care

How patients respond to medical errors may influence how physicians approach disclosure of medical errors, but information on patients' responses is limited. Research is needed on how the circumstances that surround a medical error affect how patients respond. 

To investigate whether patients' tendency to forgive a physician following a medical error varied under different circumstances. 

We mailed a questionnaire to 1500 randomly selected health plan members; the response rate was 66%. Questionnaire items assessed the likelihood of forgiveness following a medical error under 12 circumstances drawn from a review of the literature. 

Respondents were most likely to forgive a physician if the patient failed to provide complete information (93% would or might forgive) and least likely to forgive if the error was due to efforts to keep costs down (11% would or might forgive). Most respondents would not forgive a physician when the physician was tired or distracted (68%), was incomplete in data collection (76%), lacked knowledge (78%), or failed to follow up (85%). Men were more likely to forgive than women; the most educated respondents were most likely to forgive. 

Our findings suggest that patients are not likely to forgive a physician in circumstances in which they suspect incompetence, inattention, or a lack of caring on the part of the physician involved. A more comprehensive understanding of forgiveness and the effect of forgiveness on the physician-patient relationship following a medical error is needed. 

Since the publication of the Institute of Medicine's report 

the problem of medical errors has gained increased recognition by the general public.

Researchers and policy makers have begun to focus serious efforts on improving patient safety,

but the complexity and uncertainty inherent in the practice of medicine make it unlikely that medical errors will ever be eliminated completely. A growing body of literature indicates that patients and the public favor full disclosure of errors to patients.

Although physicians express support for disclosure in theory,

Physicians' reluctance to disclose errors to patients may stem in part from concern about how patients will respond to knowledge of a medical error. The objective of this study was to explore patients' attitudes toward medical errors, especially their willingness to forgive when medical errors occur. Physicians who understand patients' perspectives are likely to be better able to communicate with patients. In addition, understanding of patients' views may help in prioritizing efforts to improve patient safety and reduce medical errors. 

This study was conducted in conjunction with a larger study

of health plan members' views on disclosure of medical errors reported previously. This report follows from and builds on that earlier work. 

We mailed a questionnaire to 1500 randomly selected adult members of a New England-based mixed-model health maintenance organization. Individuals could join the health maintenance organization through an employer or via Medicaid or Medicare. The details of the mailing are reported elsewhere.

Respondents were asked whether they would forgive a physician who had made an error under 12 circumstances (

). The circumstances listed were intended to be representative of situations cited in the literature as possibly contributory to error.

Three response options were provided, including "would forgive," "might forgive," and "would not forgive." We also collected background and demographic information, including sex, age, race, ethnicity, and education. 

Background characteristics of respondents and nonrespondents were compared using a Ã·

test for age. Percentages of respondents choosing each response category for each item were calculated. A summary "tendency-to-forgive" score was computed, indicating the percentage of the 12 circumstances in which the respondent would or might forgive. 

tests to examine the relationship between respondents' sex and forgiveness. We used analysis of variance to examine the relationship between level of education and forgiveness (with education categorized as high school or less, some college or a 4-year college degree, or greater than a 4-year college degree). To examine the relationship between age and forgiveness, we used 

tests comparing those 64 years or younger with those 65 years or older; we also calculated the correlation between age and forgiveness. Statistical analyses were performed using SPSS software, version 11.5 (SPSS Inc, Chicago, Ill). The study protocol was approved by the institutional review boards of the participating institutions. 

Nine hundred fifty-eight complete and usable questionnaires were returned, for a response rate of 66% (47 undeliverable questionnaires were removed from the denominator for response rate calculations). Compared with nonrespondents, respondents were more likely to be female (57% vs 45%, 

< .001). Respondents were also older than nonrespondents (mean age, 51.9 vs 45.0 years; 

< .001). Nearly all respondents (92%) reported their race as white, consistent with the US Census statistics for this area for 2000, where 89.6% of respondents identified themselves as white.

With respect to education, 29% of respondents 25 years or older reported having a 4-year college degree or higher, compared with 27% of census respondents 25 years or older.

The Table gives the percentage of respondents choosing each response option for the 12 circumstances listed, ordered from most forgivable to least (based on the percentage responding that they would forgive). Comparing summary tendency-to-forgive scores, men were more likely to forgive than women (55.7 vs 49.5, 

< .001). Although we did not use statistical tests to make comparisons at the item level, men were more forgiving than women for each individual item. We also found a relationship between education and forgiveness, with the most educated respondents reporting the strongest tendency to forgive (F = 6.439, 

= .002); again, this tendency was consistent across individual items. The mean + SD tendency-to-forgive score for the most educated respondents was 58.4 + 22.9, compared with 51.0 + 21.3 for those with some college education or a 4-year college degree and was 51.2 + 22.5 for those with a high school education or less. We found no statistically significant relationship between tendency-to-forgive scores and age. 

To our knowledge, this study is the first to examine patients' opinions about forgiving physicians involved in medical errors under different circumstances. We found that patients consider some circumstances more forgivable than others. For each circumstance, there was one response category that most respondents endorsed, suggesting that circumstances can be considered as 3 sets based on these modal responses, including circumstances that most respondents would forgive, those that most might forgive, and those that most would not forgive. With respect to the first set, most respondents indicated that they would be forgiving of the circumstance in which the patient contributed to the error by not telling the physician everything relevant to his or her condition. This observation is consistent with previous findings that most respondents believed that patients should take some responsibility for preventing errors.

There were 6 circumstances in which most respondents indicated that they might be forgiving. Three of these circumstances (treatment of an unusual condition, in an emergency situation, or by a physician who is not the patient's regular physician) would probably be considered unusual or atypical by the public. The other 3 circumstances (a physician receiving bad advice from another physician, being too aggressive in treatment, or not being aggressive enough in treatment) suggest that the physician was at least trying to provide good care. In contrast, the last set of circumstances, which most respondents would not forgive, would probably be considered by lay people to be indicative of incompetence, inattention, or an uncaring attitude. It has been suggested that some offenses, especially those that violate assumptions about the goodness of others, may be unforgivable.

In the circumstance in which the physician's consideration of costs played a role, only 3% of participants indicated that they would forgive the physician. The public may see consideration of costs as so inimical to physicians' roles as healers that an error in this circumstance may be considered especially egregious and unforgivable. 

We found that men reported more forgiving attitudes than women. Wording of the items was intentionally gender neutral, but we have no way of knowing whether respondents made assumptions about the sex of the physician. Conventional wisdom might predict that women would be more likely to be empathetic and would therefore be more forgiving of the physician. Our findings suggest the opposite. If women are more empathetic, it appears that they may empathize with the patient rather than the physician. Our results are consistent with findings reported by Worthington,

who found that women tended to favor the patient in a fictionalized malpractice suit, while men favored the physician. 

Our finding that the most educated respondents had higher tendency-to-forgive scores may be due to more educated respondents' perceiving themselves as more similar to the physicians. It is also possible that more educated respondents are more likely to be aware that the practice of medicine is replete with uncertainty and that physicians' actions cannot be easily characterized as right or wrong. Further research is needed to determine whether either hypothesis is supported and whether this result is confirmed in other samples. 

These findings have implications for physicians who are concerned about disclosure of medical errors. One component of the disclosure process may be an apology; furthermore, in many cases an explanation is critical to the effectiveness of an apology.

of the effect of apology in a videotaped office setting where a patient had an extended wait before being seen, participants responded more positively when the apology included an explanation compared with when it did not. However, our findings suggest that not all explanations are likely to make an apology more acceptable. From the patient's perspective, at least some circumstances may not be forgivable. Patients have high expectations of physicians even in difficult circumstances, and certain explanations (eg, those having to do with avoiding cost or being rushed) are not likely to be well accepted by patients. 

This study has several limitations. Responses to questionnaire items may not correspond to responses to real occurrences of medical errors. The items used herein were abstract, which may have made it more likely that respondents would express unforgiving attitudes compared with how they would react toward their own physician, with whom they have a relationship. We are not aware of any studies that have examined patients' tendencies to forgive their own physician, so we have no empirical findings that would allow us to determine whether or to what extent our findings predict how patients would respond to their own physician under similar circumstances. We did not define forgiveness, so we are unable to determine whether respondents considered the emotional, cognitive, or behavioral correlates of forgiveness in formulating their judgments. We are not able to address the extent to which patients might experience forgiveness internally but still seek redress. In instances of actual error, there are other factors that might be expected to influence a patient's response, including, for instance, whether multiple providers were involved in the occurrence. The fact that the study was conducted within the context of a managed care plan may also have had an effect on patients' perceptions and responses, especially with respect to the item referring to errors due to efforts to keep costs down. Our ability to generalize these findings may be limited because we drew our sample from a single geographic area. Because of a lack of sufficient racial and ethnic diversity in our sample, we were unable to examine whether there are racial or ethnic differences in tendencies to forgive; additional studies examining whether differences exist will be important. Last, although we had a large sample and a response rate of 66%, nonresponse bias remains a potential limitation. 

Forgiveness is seen as an important step for physicians, patients, and families in the emotional healing that follows error and injury. It has been argued that physicians must apologize and be forgiven, after an error has occurred, before trust can be reestablished and the patient and physician can move forward.

highlights, it is important for clinicians to be aware that forgiveness of medical errors may not be, and perhaps should not be, automatic. Our findings suggest that patients are not likely to forgive a physician in circumstances in which they suspect incompetence, inattention, or a lack of caring on the part of the physician involved. More work is needed to understand how forgiveness affects the physician-patient relationship after a medical error has occurred. 

Health Plan Members' Views on Forgiving Medical Errors

 To evaluate the feasibility of using health-planadministrative data to measure potential drugâ€“drug interaction(DDI) rates in the ambulatory setting at the medical-group leveland to assess the potential use of DDI rates in performance measurement,quality improvement, and research in patient safety.

 We combined administrative and pharmacyclaims data from 2 large health plans to calculate the rates atwhich member users of selected chronic medications were potentiallyexposed to a second drug known to pose a risk of harmfulinteractions.

 We divided 44 medication combinations with risk ofadverse interactions into those with DDIs of moderate/severe clinicalsignificance and those with DDIs of mild significance. We thencalculated yearly rates of potential DDIs in continuously enrolledmembers aged 19 and older from 1998 through 2001. Rates werecalculated for all members, overall base-medication users, andindividual medical groups responsible for their care.

 The analytic data set included 756 047 patient-years ofdata and 110 to 123 medical groups per year. During the 4-yearinterval, one or more unique potential DDIs occurred in 6.2% to6.7% of base-drug users and 2.0% to 2.3% of all adult health-planmembers per year. Medical-group mean user rates were slightlylower (5.33%â€“5.81%), with wide variance (SD = 2.6%â€“3.1%) andhigh stability over time.

 Potential DDI rates calculated from health-plandata have promise for measurement in patient medication safety.This readily available and inexpensive evaluation tool has potentialfor monitoring, improvement, and research purposes if furtherstudies validate their relationship to actual adverse events.

Release of the 1999 Institute of Medicine (IOM)report on medical errors created a public furorand stimulated many research and improvementinitiatives.

 Although the IOM report recommended a50% reduction in errors over 5 years, it did not discussmeasurement of errors or adverse event rates as ameans of monitoring trends and improvement efforts.

The studies cited in the IOM report calculated errorswith methods too expensive and labor intensive forrepeated use in performance monitoring or improvementefforts. Moreover, the studies dealt with hospitalincidents only, with uncertain applicability to ambulatorycare. Thus, we know very little about the nature orepidemiology of errors or adverse events in ambulatorycare settings, and we lack measurement methods foraccountability or improvement.

In the absence of any practical methods of quantifyingoverall errors, measurement of adverse drug events(ADEs) may be a good way to gauge errors and patientsafety in ambulatory care, since studies have found ADEsto be frequent and a significant contributor to emergencydepartment visits and hospitalizations.

 Using multiplemethods in 2003, Gurwitz found an incidence of 50.1ADEs per 1000 person-years in those aged 65; 27.6% weredeemed preventable.

 Gandhi concluded from a patientsurvey that during a 3-month period, 25% of outpatientswith a prescription experienced an ADE, 13% of whichwere serious and 11% preventable.

 Both studies arevaluable in understanding the approximate rates ofADEs but neither is amenable to repeated measures atthe level of individual practices or medical groups.

Although ADEs appear to be relatively frequent,there are practical limitations in relying on chart auditsand surveys to detect and monitor them. Drug claimdata and member information available from healthplans offer an inexpensive and repeatable way to measurepotential ADEs, especially drug—drug interactions(DDIs), in ambulatory care. A DDI is "2 or more drugsinteracting in such a manner that the effectiveness ortoxicity of 1 or more drugs is altered."

 A potential DDIshould be identifiable from health-plan claims data thatshow 2 such drugs prescribed simultaneously.Researchers have studied DDI rates in special populationsand settings,

 but only 2 recent studies haveused large population-based pharmacy claims databasesto study case rates of specific drug pairs. Peng et al useda drug utilization review program to study potentialDDIs in a population of 3 million, and Hennessy et al didthe same for enrollees in 6 state Medicaid programs.

However, no one has studied DDI rates in individualmedical groups, the level at which accountability andimprovement ability reside. Such analysis can facilitatemeasurement and comparison of performance andprovide medical groups with a basis for qualityimprovement.

Therefore, we combined administrative data from 2large health plans to answer the following questions:

 Is it feasible to use health-plan data to identifypotential DDI rates in relation to the individualmedical groups responsible for member care?

 Is the frequency of potential DDIs high enough tocharacterize medical groups or clinics?

 How much variation in rates exists among medicalgroups?

 Are these rates stable enough over time to serve asa credible way to measure trends and changes?

We studied the frequency of potential DDIs from1998 through 2001 in health-plan members served bythe medical groups contracting with 2 HMOs. The primarydata sources were pharmacy claims, medical careclaims, and membership files. The institutional reviewboards at both organizations approved the study.

The HMOs participating in this study were HealthPartnersin Minnesota and Lovelace Health Plan in NewMexico. In 2000, HealthPartners covered 657 000 membersthrough contracts with 50 medical groups, andLovelace Health Plan had 240 000 members cared for byapproximately 240 medical groups.

The study included all medical groups that providedprimary care for at least 30 health-plan members from1998 through 2001. Location, based on US Bureau ofthe Census designation, and other descriptive dataabout these medical groups were collected from health-planrecords.

We identified HMO members who were 19 or older asof January 1 in the study year, had continuous membership(with up to a 45-day break) with pharmacy benefitsduring the study year, and had a known affiliationwith an eligible medical group. Descriptive data aboutmembers were obtained from membership and claimsrecords.

We included in our analysis 44 combinations of drugs(a base drug and a conflicting drug or drugs) known tohave potential adverse interactions and in which thebase drug is taken chronically by large numbers of people

). We used 3 key references to identify druginteraction combinations: (1) Hansten and Horn's DrugInteractions, Analysis and Management, (2) the DRUGREAX&#174;System of MICROMEDEX, and (3) Evaluationsof Drug Interactions™ (EDI).

 Because there is somuch variation among standard texts about what combinationsrepresent a DDI of what severity, we requiredthat each DDI combination be identified as a problem inat least 2 of these 3 references and have a clinical significancerating of 1, 2, or 3 (on a scale of 1 to 5), asassigned by Hansten and Horn.

 Severe interaction ("Avoid administration of thecombination. The risk of adverse patient outcomeprecludes concomitant administration.")

 Moderate interaction ("Avoid administrationunless it is determined that the benefit of coadministrationoutweighs the risk to the patient.Patients should be monitored closely.")

 Mild interaction ("Minimize risk by consideringalternative agents or change dosage or route ofadministration. Patient monitoring is suggested.")

Margaret J. Gunter, PhD

Articles

Colorectal Cancer Screening Use Among Insured Primary Care Patients

Monitoring of Drugs With a Narrow Therapeutic Range in Ambulatory Care

Health Plan Members' Views on Forgiving Medical Errors

Measuring Patient Safety in Ambulatory Care: Potential for Identifying Medical Group Drug"Drug Interaction Rates Using Claims Data

The Role of the ECHO Model in Outcomes Research and Clinical Practice Improvement