We have developed an automated, practical approach to collecting point-of-care information from clinicians via ATHENA-HTN. Analysis of this feedback revealed a variety of patient-, clinician-, and data-related reasons for why clinicians did not follow guidelines to intensify treatment. For many visits, we found that clinician feedback described influential patient and data factors that were not systematically captured by the electronic medical record (EMR) data available to the DSS. This suggests that many decisions currently identified by the DSS as guideline nonadherent may in fact represent clinically appropriate decisions.
Our findings shed light on several specific areas where automated decision support systems for hypertension management may be improved. Compared with human review of medical records, computer-aided review can produce different judgments about whether a clinician behavior is guideline nonadherent. This is due in part to the presence of clinically relevant information in the EMR that is entered in non-standardized formats (e.g., free-text comments).14
In a randomized trial comparing office-BP-driven hypertension treatment decisions with home-BP-driven decisions, reliance on home BPs (i.e., information often captured only in free-text in medical records) led to less intensive drug treatment.15
Our finding that clinicians frequently did not consider the recorded office BP to be representative of the patient’s typical BP is consistent with that of Ferrari et al., who found that satisfactory patient self-measured BPs were a main reason clinicians reported for not intensifying therapy.5
Standardized collection and coding of data on home BPs may improve the usefulness and applicability of displayed recommendations in facilitating optimal hypertension management. Analysis of point-of-care feedback also identified instances where the primary care clinician was not responsible for managing the patient’s hypertension. Approaches to identify the correct clinician decision-maker are needed to ensure effective guideline implementation and to appropriately target decision support.
Other factors identified in our study, such as clinician reports that hypertension was not a clinical priority or that more monitoring was needed, have been reported in previous studies.3–5
These factors may describe appropriate clinician guideline non-adherence in certain patient scenarios. However, Oliveria et al. found that clinicians often deferred intensification even when uncontrolled hypertension had been documented for at least six months preceding the patient’s most recent visit.4
Our study has several limitations. Given the extra work required to offer optional comments, it may be expected that busy clinicians would do so infrequently. Clinicians voluntarily provided feedback at 7% of visits. Many barriers may be under-reported. While provision of feedback was not associated with patient demographic characteristics or the number of active antihypertensive medications, clinicians in our study were more likely to provide feedback at visits where the patient had a significantly elevated BP and/or an elevated BP at their previous visit (data not shown). It is possible that clinicians selectively offered comments for scenarios they encountered repeatedly or for scenarios they believed required justification. We are not able to confirm this using our data. For these reasons, our findings may not be generalizable beyond our study sample. Nevertheless, these spontaneously offered comments, provided during actual patient encounters, offer additional insight into the various factors and challenges clinicians consider at the time of the medical decision.
In summary, point-of-care feedback collected from clinicians via ATHENA-HTN provides a rich source of information about perceived barriers and other factors related to adherence to guidelines. Integration of a variety of feedback features into ATHENA-HTN offered clinicians the opportunity to provide holistic impressions of the applicability of hypertension management guidelines to specific patient scenarios, as well as responses to specific drug recommendations. This ability to link data available in the EMR with computer-generated advice and clinician responses allowed us to identify several important clinician, patient, and data barriers to guideline-concordant intensification of therapy. Our study suggests that clinicians’ failure to intensify therapy may reflect valid concerns about whether this action is appropriate for many patients.