The relative accuracy of documentation of psychiatric comorbidities used in biomedical research is rarely addressed. To our knowledge, our study is among the first to contrast the content of patient reported outcomes (PROs) data with traditional medical record diagnoses, which have traditionally been used in HIV clinical cohort studies. These psychiatric comorbidities have been linked to decreased ART adherence in the past and have been found to have important implications for morbidity and mortality. Our research suggests that PROs present an alternative method of capturing data on these domains. Indeed, PROs may provide additional benefits to provider capture because they call for updates of current
status on each condition. Further, the decreased influence of social desirability bias may improve the reliability of data capture for sensitive domains such as substance use. PROs also offer the opportunity for intervention at clinical presentation because providers or other members of the health care team can be automatically alerted of issues detected by PROs. We have done so with the presence of suicidal ideation [15
When comparing data captured through EHR to data from PROs, we found significant differences in the prevalence of substance use, depression, tobacco use, and at-risk alcohol use among HIV-infected patients in routine clinical care. These differences were found despite standardized provider training in the use of EHR and an additional layer of note review by a third party, routinely used for data entry quality control at our site. Provider, patient, and structural factors are likely to influence data capture across these domains in routine care. At the provider level, longstanding diagnoses such as depression and substance use may not undergo thorough reevaluation at every visit owing to time constraints or focus being distracted by competing clinical events. Thus, longstanding diagnoses may persist in active problem lists. At the patient level, some individuals may be uneasy discussing certain topics directly with their provider, resulting in incomplete data capture. Studies have shown that data capture for sensitive domains (substance use, sexual risks, and so on) is done more effectively by PROs than healthcare providers. In part due to social desirability bias, many patients report more truthful responses to computer-administered surveys than to face-to-face interviews across sensitive domains [4
]. PRO assessments offer added value, revealing the state of the condition in real time and offering the provider an opportunity for intervention. Finally, on a structural level, PROs offer a more consistent and dynamically updated data capture platform that, if well integrated into clinic workflow, can provide efficient and standardized data capture. A benefit of such a system is that it is patient-dependent and would remain relatively unaffected by daily provider workload fluctuations and other forces that currently affect data capture at the point of care (eg, number of scheduled patients for a given clinic).
We also found significant differences in the association of these findings with poor ART adherence. The relationship between minority race, substance use, depression, and poor ART adherence has been extensively documented and well established in the HIV/AIDS literature in many prospective and retrospective trials [1
]. While an increased risk for poor ART adherence for minority race and uncontrolled viral loads was found in both EHR and PRO models, the well-known associations between substance use, depression, and poor adherence were only found in the model utilizing self-reported PRO data. In essence, the model that employed PRO-captured variables for these domains corresponded with the extant literature.
We posit the concordance of the PRO model with the known risk factors for poor ART adherence suggests increased accuracy of these data across specific domains; it also suggests that the PROs present an opportunity to enhance the quality of data elements available for comparative effectiveness research and patient care. PROs provide an adjunct screening measure for data capture at the point of care for use in clinical practice to address barriers to ART adherence. While PRO data was immediately available to clinicians, it had not been fully implemented to routine care at the time of this study. To fully realize the benefits of PRO data capture, operational and implementation research must be conducted to find ways to integrate the capture of such instruments routinely in outpatient and potentially inpatient practice settings.
Our findings should be interpreted with respect to the limitations of our study. As an observational study from a single HIV cohort, our findings may not be applicable to other national or international settings, although our analysis may provide insights applicable to such settings. As with all observational studies, we are able to identify associations but cannot attribute causality. Although we controlled for measured confounders using multivariable models, potential for unmeasured confounding inherent to all observational studies persists. In another potential limitation, PROs at the 1917 Clinic are administered by a touch-screen system without an audio counterpart. Although assistance was provided to patients with visual impairment or inadequate literacy, these factors were not controlled for in this analysis and there is not a clinic-wide measure of patient literacy at this time. In addition, we are not able to directly assess the respective diagnostic accuracy of the EHR and PRO data through comparison to a third source; rather, we used associations of well established psychiatric comorbidities with poor ART adherence as indirect evidence of improved data quality and accuracy with PROs. Further research using structured clinical interviews as a gold standard to compare data from EHR and PROs may help to evaluate the relative accuracy of PRO findings.
In conclusion, our findings suggest that PROs represent an important resource to HIV/AIDS clinics at both the level of clinical care and cohort research. Ultimately, integrating PROs at the point of care in a routine clinical setting may represent a transformative health informatics for routine clinical care and comparative effectiveness research.