The Role of Clinical Uncertainty in Treatment Decisions for Diabetic Patients with Uncontrolled Blood Pressure FREE

—The Editors

Despite some recent improvements in blood pressure control, the number of patients with inadequate control remains high and contributes to excess morbidity and mortality, especially among patients at high risk from complications of hypertension (1 - 8). Several studies have suggested that “clinical inertia”—the failure by providers to initiate or intensify therapy (medication intensification) in the face of apparent need to do so—is a main contributor to poor control of hypertension (9 - 12).

Although the failure to intensify treatment medications for patients with elevated blood pressures at visits has been well documented ((5 - 6), (12 - 18)), factors underlying what seems to be clinical inertia have been studied less systematically. When providers are queried after clinic visits about the lack of medication intensification for elevated blood pressure, they variously report that the patient's “true” blood pressure was lower than the clinic blood pressure reading, that other patient concerns precluded attention on blood pressure management, and that patient adherence should be improved before medication intensification ((6), (17)). Some studies have examined the role of various clinical and patient factors in intensification decisions ((6), (8), (17), (19 - 20)), but no study has used a detailed conceptual model to comprehensively examine the relative contribution of a broad array of potential patient, provider, organizational, and visit-specific contributors to a medication intensification decision. In addition, although a frequently cited reason for deferring medication changes is that the clinic blood pressure does not reflect the patient's “true” blood pressure (21 - 22), this clinical uncertainty and its effects have not been explored.

To better understand factors underlying apparent clinical inertia for hypertension, we designed the ABATe (Addressing Barriers to Treatment for Hypertension) study to examine treatment change decisions for diabetic primary care patients with elevated triage blood pressures before a primary care visit. We defined elevated blood pressure for this population to be 140/90 mm Hg, a value well above guideline targets for diabetic patients and one clearly requiring some type of action (4). Our goals were to assess how often patients presenting with an elevated triage blood pressure received medication intensification or were scheduled for close follow-up and the role that clinical uncertainty about blood pressure, competing demands and prioritization, medication-related factors, and care organization play in treatment change decisions.

On the basis of theories of patient, provider, and organization behavior (23 - 36), we developed a conceptual model—the hypertension clinical action model—to examine decisions that drive treatment change (medication intensification or prompt blood pressure follow-up) for elevated blood pressure (Figure 1). The model, developed by 2 internists and 3 PhD-level methodologists in conjunction with development of ABATe and before data collection, proposes such treatment change decisions at a visit are based on 4 main conceptual domains: clinical uncertainty (Is the patient's blood pressure truly elevated? Does the clinic blood pressure reflect the “true” blood pressure?), competing demands and prioritization (What other problems need to be addressed at this visit? Is blood pressure management a priority for this particular patient? Does the provider place priority on blood pressure management in general?), medication-related factors (Should adherence be addressed first? Is the medication regimen too complex? Will the patient accept another medication?), and care organization (Is there sufficient time to address hypertension? Are staff available for follow-up?). In addition, as part of grant development, we hypothesized that the following factors would lead to a lower probability of treatment change: uncertainty about whether the patient's visit blood pressure reflected their true blood pressure (“clinical uncertainty”), comorbid conditions unrelated to hypertension and diabetes (37), a lower priority placed by the provider on the importance of treating elevated blood pressure, a higher number of baseline medications, perceived medication adherence problems, shorter clinic visit times, and lack of staff to follow up for blood pressure medication adjustment.

We conducted a prospective cohort study of patients with scheduled primary care visits at 9 Veterans Affairs facilities located in 3 midwestern states. These facilities varied in size and structure, with 3 large academic-affiliated medical centers, 1 large nonacademic medical center, and 1 large and 4 small community-based outpatient clinics. From 15 February 2005 to 14 February 2006, approximately 33 500 diabetic patients visited primary care providers (including residents) in these facilities (range per facility, 1050 to 9200 diabetic patients). The institutional review boards of all participating facilities approved the study protocol. Both patients and providers gave written informed consent before participating. Providers received a $50 bookstore gift card, and patients received a $10 department store gift card for completing initial surveys. Providers were told that the study was about diabetes and hypertension, with the purpose being to “study challenges in treating patients with diabetes and ways to overcome these challenges so that quality of care can be enhanced.”

We invited all nonresident primary care providers with patient care responsibility at least 2 half-days per week to participate in the study. Of the eligible 126 providers approached, 104 consented to participate, for an overall recruitment rate of 83% (median facility-level recruitment rate, 88%). By the time recruitment started, 12 providers had stopped working at their facility or changed their patient care responsibilities, leaving 92 primary care providers still eligible to participate (range per facility, 2 to 28 providers; median, 8).

As specified by our institutional review board protocols, potentially eligible patients were referred to study staff by triage personnel. During the enrollment periods at each facility, study staff screened all referred patients who presented for a scheduled visit to participating primary care providers and whose lowest triage systolic blood pressure was 140 mm Hg or greater or whose lowest triage diastolic blood pressure was 90 mm Hg or greater (Figure 2). In each of the facilities, triage staff routinely used an electronic cuff to check the patient's blood pressure before the provider visit. Triage policies specified that a second blood pressure measurement should be obtained if the first blood pressure was elevated. In addition to the triage blood pressure, study staff screened patients for the following inclusion criteria: the participant confirmed a diagnosis of diabetes, the participating provider was the primary provider of diabetes care for the participant, and the participant spoke English. Patients with impaired decision-making ability (for example, dementia and traumatic brain injury) or terminal disease and residents of nursing homes were excluded. Of the 1556 patients approached by study staff, 213 were ineligible (Figure 2) and 1169 provided written informed consent to participate in the study (approached and eligible, 87%; median facility-level recruitment rate, 89%). We enrolled a median of 14 patients per provider (range, 1 to 16 patients) from February 2005 to March 2006. Recruitment time per facility varied from 4 to 12 months.

Our prespecified sample size calculations stipulated that we needed at least 11 patients from 80 physicians across 8 sites (that is, 880 patients) to detect a moderate difference in treatment change (about 12%) when competing demands were or were not present.

We included data from 5 sources in our analysis (Table 1). First, a baseline survey completed by all providers provided variables assessing provider prioritization to blood pressure management, general provider characteristics, and availability of ancillary support for blood pressure management. Second, providers completed a brief visit survey for each patient after the same clinic session in which they saw the patient (completion rate, 99%). This survey provided information on which issues were discussed during the visit, the provider's blood pressure goal for the patient, and whether medications were changed during the visit. Third, a patient survey conducted at enrollment provided sociodemographic characteristics, self-reported adherence and difficulty with medications, and self-management practices (completion rate, 91%). Fourth, review of electronic medical records documented free text blood pressure values and notes on actions taken at the enrollment visit. Finally, patient age, prescribed medications and their dosages, triage blood pressure values, and comorbid conditions (International Classification of Diseases, Ninth Revision, codes in the previous year) were obtained from Veterans Health Administration automated data sources.

Our main dependent variable was whether a provider changed a patient's blood pressure treatment at the visit in response to the elevated triage blood pressure. We considered treatment change to have occurred if the provider indicated on the visit survey (or in the medical record if a visit survey was not completed) that he or she added a medication or increased the dose of an existing medication or if the provider documented in the medical record that he or she intended to have the patient return for blood pressure follow-up within 4 weeks. We wanted to capture whether the provider determined that blood pressure was of sufficient priority at the visit to either intensify treatment or to specifically note a need for prompt follow-up. We considered only documented plans for follow-up because medical record documentation is currently the standard for establishing treatment plans and the main way to communicate to other providers that blood pressure at that visit was a problem that needed to be addressed.

Table 1 summarizes the independent variables, categorized according to the dimensions in our conceptual model and their data sources. Covariates related to blood pressure control included the visit systolic and diastolic blood pressures and the mean systolic blood pressure in the previous year (calculated as the mean of all systolic blood pressures in the previous year stored in the automated data).

We assessed competing comorbid demands in 2 ways. First, we categorized all conditions reported by providers to have been discussed during the visit as either related (concordant) or unrelated (discordant) to diabetes and hypertension (37). Concordant conditions included hyperlipidemia, obesity, heart failure, ischemic heart disease, peripheral vascular disease, renal disease, and cerebrovascular disease. All other conditions were considered discordant. Second, we used the method described by the Department of Veterans Affairs Health Economics Resource Center (38) to identify chronic conditions prevalent in our study sample during the year before the visit by using codes from the International Classification of Diseases, Ninth Revision. We similarly classified these conditions as concordant or discordant.

We first examined associations between treatment change and visit systolic blood pressure, visit diastolic blood pressure, and systolic blood pressure in the previous year by using a 3-level logistic regression model; the first level addressed patient variables, the second level addressed the primary care provider, and the third level addressed the site where the provider worked. Multilevel models allow us to explicitly estimate the variability in treatment change attributable to provider and site while appropriately accounting for the clustering of patients within providers and sites. Next, we separately examined the association of each independent variable with treatment change. Because it makes sense clinically that blood pressure should be the principal determinant of treatment change decisions, we controlled for visit systolic blood pressure, visit diastolic blood pressure, and mean systolic blood pressure in the previous year in all of our models (referred to as “blood pressure–adjusted models”).

We then constructed a multilevel model (referred to as the “final multivariate model”) with all the independent variables that were associated with treatment change (P ≤ 0.20) in the analyses above (we also required a monotonic pattern of response if added as a set of ordinal dummy variables). Variables were then eliminated in a backward fashion if they were no longer statistically significant (P < 0.05). We assessed model fit for variables or groups of variables by using a likelihood ratio test between nested models. Estimates of the proportion of variance explained by predictors and variance components are for the latent variable represented by the logistic regression (41).

For clarity of presentation, we calculated a probability of treatment change from both the blood pressure–adjusted and final multivariate models across levels of each independent variable for the modal provider and a patient with visit systolic, diastolic, and mean previous year blood pressures set at the population means. The presented results are from the final multivariate model.

The missing data rates were low for variables not collected by survey, ranging from 0% to 1.6%. The patient survey had a completion rate of 91%, and the variables used from the survey had a missing data rate of 9% to 15.5% (for income). The blood pressure–adjusted model was conducted for the participants with nonmissing data, and the numbers for each variable are reported. In the final multivariate model, no survey variables entered, and only 68 of 1169 patients had any missing values.

All analyses were conducted by using STATA software, version 10.0, with the xtmelogit procedure (Stata, College Station, Texas).

This study was funded by the U.S. Department of Veterans Affairs Health Services Research and Development Service. The salaries of 2 of the authors were also supported in part by the Michigan Diabetes Research and Training Center Grant (P60DK-20572) from the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health. The funding sources were not involved in the conduct or analysis of this study.

The mean age of the 1169 participants was 66 years (SD, 11), and 80% were white. On average, patients reported having had diabetes for 11 years. The mean age of the 92 primary care providers was 48 years (SD, 9). Among the providers, 64 were physicians, 21 were nurse practitioners, and 7 were physician assistants. Tables 2 and 3 list other baseline patient and provider characteristics.

Overall, 573 of 1169 (49%) patients had a treatment change at the visit. Of those, 511 received an initial dose of medication or had a dose increased and 62 had no medication intensification but had a documented plan to follow up blood pressure within 4 weeks. The mean triage systolic blood pressure at enrollment was 154 mm Hg (SD, 14), and the mean triage diastolic blood pressure was 78 mm Hg(SD, 12). The mean previous year systolic blood pressure was 145 mm Hg (SD, 15). The likelihood of treatment change increased substantially with higher systolic and diastolic blood pressures at triage and higher systolic blood pressure during the previous year (Figure 3). There were no statistically significant differences in treatment change associated with patient age, education, or race or provider age (data not shown).

Patient report that they self-monitored blood pressure at home was not associated with treatment change (Table 3); however, when providers recorded at the visit that the patient's blood pressure as measured at home was adequate, providers were less likely to change treatment (18% vs. 52%; P < 0.001). Similarly, as expected, treatment was much less likely to be changed when a provider recorded a repeated blood pressure measurement during the visit as less than 140/90 mm Hg than when repeated measurement was 140/90 mm Hg or greater or when no repeated blood pressure was recorded (13% vs. 61%; P < 0.001).

When providers reported discussing a condition that was discordant with diabetes and hypertension (for example, chronic pain or preventive care) during the visit, they were less likely to initiate treatment change at that visit (Table 3). However, the total number of discordant chronic conditions in the past year was not associated with treatment change as a continuous variable (although the odds of treatment change were decreased for the small group with ≥5 discordant chronic conditions).

Providers whose systolic blood pressure goal for a given patient was 130 mm Hg or less were more likely than those with higher treatment goals to initiate treatment change (52% vs. 33%; P < 0.002). Providers who reported willingness to wait longer than 4 weeks before following up a modestly elevated blood pressure on the baseline survey were much less likely to initiate treatment change at the visit than those willing to wait 2 weeks or less (41% vs. 58%; P < 0.03), but providers' general propensity to intensify blood pressure medications was not associated with treatment change (Table 3).

Contrary to our hypotheses, the number of antihypertensive medications, dosing, and the total number of medications were not statistically significantly associated with treatment change (Table 3). Treatment change was much less likely to occur when the provider reported that he or she discussed medication adherence or other medication problems at the visit, compared with providers who did not discuss such issues (23% vs. 52%; P < 0.001). However, patient self-reported medication adherence was not associated with treatment change, nor was patient self-reported difficulty with adding a new blood pressure medication.

The average rate of intensification varied greatly by facility (26% to 65%; P = 0.033 for the site coefficients as fixed effects). However, contrary to our hypotheses, the provider reported that the average number of patients seen in a half-day clinic, the minutes allotted for a return visit, and the available staff to follow-up high blood pressure were not statistically significantly associated with treatment change.

In the final 3-level model, the 3 blood pressure covariates (visit systolic blood pressure, visit diastolic blood pressure, and previous year systolic blood pressure) together explained 12% of variance in treatment changes (Appendix Table). The variables reflecting clinical uncertainty at the visit (repeated blood pressure <140 mm Hg and adequate home blood pressure measurements) were strongly associated with decreased odds of treatment change (odds ratio for treatment change, 0.09 and 0.20, respectively) and together increased the variance explained by the predictors from 12% to 29%. For the full model, the independent predictors together explained 35% of the variance in treatment change. In addition, a higher provider systolic blood pressure goal, a threshold for follow-up greater than 4 weeks, and a discussion about medication problems or discordant conditions at the visit were strongly associated with decreased odds of treatment change (odds ratios, 0.27 to 0.66).

Because the clinical uncertainty variables were so highly predictive of treatment change, we examined the variability in rates of these 2 variables across providers and facilities. Overall, providers recorded a repeated blood pressure value in 529 (45%) visits. The provider rates of repeating blood pressure measurements varied considerably (interquartile range, 11% to 80%). Recording a repeated blood pressure value varied enormously by site, from 16% of encounters to 95% of encounters. Similar provider and site variability was seen in the documentation of adequate home blood pressure measurements, although the overall rate of such documentation was much lower (about 8% of encounters).

For approximately 50% of diabetic patients in our study who presented with a triage blood pressure of 140/90 mm Hg or greater, providers initiated treatment change (that is, intensified medication or planned close follow-up) at the time of the primary care visit. Even after controlling for several other predictors, factors related to clinical uncertainty at the visit about the true blood pressure value were the most prominent predictors of treatment change. Specifically, when primary care providers recorded lower repeated blood pressure assessments and considered patient reports of lower home blood pressure values, they were much less likely to initiate treatment change. Competing demands, in contrast, impeded treatment change more modestly and primarily, as others have found (20), when the comorbid conditions were specifically addressed at the visit. This suggests that comorbid conditions act as competing demands not simply because of their number but rather because of their severity, acuity, or dominance at a given point in time ((37), (42)).

Two other factors related to blood pressure prioritization were related to treatment change. Specifically, providers with lower blood pressure treatment goals and those with shorter general thresholds for follow-up of an elevated blood pressure reading were more likely to initiate treatment change. This finding is important because such factors are potentially modifiable by educational efforts.

We purposely chose to evaluate repeated blood pressure assessments by providers independent of the triage measurements and categorize them as reflecting “clinical uncertainty.” Given natural variation in blood pressure (43) and differences in techniques to assess blood pressure (44), providers have often been trained to recheck an elevated reading (usually by using an auscultatory method), with the implication that they should trust the value they obtain more than the value obtained at triage (usually obtained by using an electronic cuff). Thus, repeating the blood pressure measurement reflects uncertainty about whether the elevated triage reading accurately reflects the patient's true blood pressure. However, although providers often treat the repeated blood pressure as the gold standard, evidence indicates that provider values are more susceptible to bias than automated office or home blood pressure readings because of terminal digit preference (rounding to the nearest 10), threshold bias (rounding below a treatment threshold), and treatment bias (expecting a lower blood pressure in those receiving treatment) (45 - 47). In addition, other studies have demonstrated statistically significant variation among physicians in the magnitude of differences between triage blood pressures measured by a nurse using an electronic method and those measured by physicians using auscultation (48) and large interobserver variation in blood pressure values using auscultatory methods (49).

Furthermore, in our study, primary care providers' decisions about when to repeat a blood pressure were not systematic—some providers repeated blood pressure readings consistently, whereas others did so only occasionally. Much confusion exists about which blood pressure value or which combination of values would best represent a patient's true blood pressure. Even the Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (50) is not clear on this point. Although the complete report (50) suggests that an average of 2 values should be used for decision making, the summary report simply states that 2 measurements should be made but not whether an average or the lower of the 2 values should be used (4). Nonetheless, our results show that when the provider repeated a patient's blood pressure measurement and the result was less than 140/90 mm Hg, providers almost always acted as if this repeated value more accurately reflected the true blood pressure and did not initiate medication intensification or schedule blood pressure follow-up. Providers in our study were also much less likely to initiate treatment change if patients reported “adequate” home blood pressures; there is a similar lack of clarity about incorporating home blood pressure reports in clinical practice ((4), (51 - 52)).

To our knowledge, ours is the largest study with the most clinically detailed data sources to systematically examine potential reasons for clinical inertia in hypertension management. Although many English-language, peer-reviewed studies from 1995 to present have documented potential clinical or therapeutic inertia for patients with hypertension ((12), (14 - 18), (21 - 22), (53 - 55)), none has systematically elucidated the role of clinical uncertainty in medication intensification decisions. Nonetheless, some investigators have noted that the reasons providers do not change medications often has to do with the belief that the clinic blood pressure does not reflect the patient's true blood pressure (21 - 22). Safford and colleagues (56) developed a cognitive map of reasons why physicians would not intensify medications for a hypothetical patient with elevated blood pressure. Among the many factors offered, the role of “white coat hypertension” and “good ambulatory blood pressure” was enumerated as contributing to reasons for clinical inaction.

Several limitations of this study should be noted. Medication intensification rates are typically higher in Veterans Affairs than in non–Veterans Affairs settings, and in fact, the intensification rate that we saw was higher than that reported in both Veterans Affairs ((5), (21)) and non–Veterans Affairs populations ((6), (12)). This higher rate may stem from our definition of treatment change, secular trends, and the fact that providers knew that they were participating in a study about diabetes and hypertension (although they were not aware of study hypotheses or which patients were enrolled in the study until after the visit). However, a strength of our multisite design is that financial access to medications among Veterans Affairs patients is relatively homogeneous; thus, differences among patients in the ability to pay is unlikely to contribute to site variation in decision making about their blood pressure treatments. Our study was designed to examine cross-sectional treatment change: that is, change measured at the time of 1 visit. The 50% treatment change rate at any 1 visit may translate to high levels of intensification over time, but the focus on a single visit allowed us to better identify patient, provider, organizational, and visit predictors that may be diluted in longitudinal studies. In addition, our treatment change variable incorporated follow-up plans documented in the medical record. Providers may have asked patients to follow up without documenting the plan, and even when follow-up was planned, it may have not always been completed. Therefore, integrating our results with information about factors that serve as barriers to medication intensification over time, including completion of follow-up, is needed.

In conclusion, by using a detailed conceptual model and data reports from providers, patients, and the medical record, we delineated factors related to clinical uncertainty, competing demands, and provider blood pressure prioritization and what influenced treatment change decisions for high-risk patients with elevated blood pressure. Perhaps our most actionable finding is that rather than simply failing to act (inertia), providers are often confronted with the inherent clinical uncertainty about blood pressure values and document actions to incorporate additional information (for example, repeating measurements or eliciting home blood pressure values), which in turn has an enormous effect on decisions to change treatment. Unfortunately, providers are doing so without a systematic approach and possibly placing undue faith in their own repeated measurements or home blood pressure values. Although providers' reliance on these additional measures might be considered a way to justify lack of medication intensification through “soft” reasons or “wishful thinking” (9), it also seems to represent true clinical uncertainty about which values to believe. Regardless of what we call it, it is clear that this ambiguous approach to using blood pressure measurements is fraught with potential bias that could undermine performance improvement initiatives and may be a major obstacle to optimizing management of hypertension and improving outcomes for high-risk populations. We must promote more systematic approaches to the use of clinic and home blood pressure measurements in the treatment of hypertension.