Counter to our hypothesis, EMRs were associated with half the odds of depression treatment including antidepressant medication and/or mental health counseling in visits by patients with three or more chronic conditions, while EMRs were not associated with receipt of depression treatment including antidepressant medication in visits by patients with two or fewer chronic conditions.
Carefully conducted studies have demonstrated that EMRs encourage biomedical exchange between the physician and patient including discussion of medication.23–27
In contrast, EMRs have been observed to have a negative impact on psychosocial exchange, with screen gaze being inversely related to physician engagement in psychosocial questioning and emotional responsiveness.24,25,28,29
It is possible that the clinical workflows embedded in EMRs inadvertently encourage physicians to focus on these multiple physical problems and push depression treatment “off the radar screen” even after physicians diagnosed the condition. Implementation and use of health information technology typically involves significant changes to clinical processes and workflows, which can have unintended positive and negative effects on care quality.17–19,30
While most prior research has been conducted in inpatient settings, it has shown that physicians often find that EMR interfaces create additional work by forcing them to click through many screens and options as well as imposing tasks previously handled by others, especially when placing orders.31,32
Similar effects in primary care may take away significant visit time and reduce physician’s cognitive performance in terms of ability to provide comprehensive care. Such effects are also likely to be significantly greater during visits by patients with multiple chronic conditions than patients with few chronic conditions.
While the relationship we observe may be attributable to EMR impact on physician–patient interaction, the study’s non-experimental design does not allow for causal inference and makes it important to consider other competing explanations for the relationships observed, particularly differences between EMR and non-EMR practices in patients, physicians, and other practice characteristics. It is possible that patients with multiple chronic conditions who do not want depression treatment may self-select into practices with a “high tech” focus, whereas patients who have yet to develop chronic conditions may be more open to both “high tech” practices and to depression treatment. Similarly, patients whose diagnosed depression is more severe may be more likely to select practices without EMRs. While there is no way to evaluate this possibility, we suspect that self-selection cannot account for the sizable effects we report in comorbid patients given the complex considerations that influence patient selection of physicians. Additionally, the analysis controls for type of insurance, income, and past utilization of office visits, which should help minimize any potential bias.
It is also possible that primary care physicians who do not want to provide depression treatment may self-select into practices with a “high tech” focus; however, it is hard to explain why these physicians would deliberately identify their patient’s depression if they had no intention to treat it, making this type of selection unlikely. Additionally, the analyses control for practice ownership, which should capture some differences in “tech” focus. Lastly, it is possible that other characteristics of the practice which co-vary with EMR use are the actual causal factor. We think other practice characteristics are an unlikely explanation for the relationships we observed because these practice characteristics that co-vary with EMR use should have influenced depression treatment of all depressed patients in the practice, not just depressed patients with multiple chronic conditions.
The internal and external validity of our findings is subject to the following considerations. First, the diagnostic accuracy of the sample is imperfect because it relies on physician judgment rather than objective assessment tools. While policy analysts find it useful to generalize to patients who receive “real world” diagnoses, this potential measurement error is problematic if diagnostic accuracy differs by the availability of EMRs. Second, the database does not contain a comprehensive set of clinical covariates so we can only hypothesize that the differences we observe by the number of chronic condition is reflective of varying clinical complexity. Third, since the unit of analysis is a single office visit, frequently visiting patients with potentially greater severity may be over-represented; however given that each office sampled visits for a one-week period only, this source of measurement error is not likely to greatly bias findings. Even with these limitations, the database is the most comprehensive national survey of EMR use by office-based physician visits over geography, population, and time.
Although this study cannot identify the exact reasons why depression treatment is less likely during visits to EMR practices than non-EMR practices, this study should raise questions about a potential downside of EMR use. EMR use involves significant changes to clinical processes and workflows compared to paper-based medical care. These changes need to be well understood in order to guard against unintended negative consequences. While EMRs certainly have advantages within primary care settings, they may result in encouraging physicians to focus on issues identified by the EMR rather than those raised by the patient, necessitating EMR re-design. Physician training on EMR and the systematic incorporation of depression treatment guidelines into EMR systems may also help to address unintended consequences we observed. EMRs require additional study to identify the extent and cause of the negative association between EMRs and depression treatment we observed, especially as more and more practices implement EMRs.