Predictors of Adverse Events and Medical Errors Among Adult Inpatients of Psychiatric Units of Acute Care General Hospitals

This study included medical records of patients discharged from psychiatric inpatient units at 14 acute care general hospitals in Pennsylvania during 2010. The Pennsylvania Health Care Cost Containment Council (PHC4) provided a data set from which a stratified random sample of inpatients from psychiatric units in general hospital settings was drawn. The PHC4 data included detailed information about patient demographic and clinical characteristics (for example, diagnosis and procedure codes, length of stay, and hospital information) for all hospital discharges in the state. From these data, a random sample of 19 general hospitals was selected, stratified by teaching status and hospital location (urban or rural), with probabilities proportional to each hospital’s number of discharges. Next, a subsample of approximately 300 patient discharges was drawn from each selected hospital, with probabilities inversely proportional to the size of the hospital. Long-term admissions with a length of stay of more than 90 days were excluded from the sample. Because five hospitals declined to participate, our total sample included 14 inpatient psychiatric units, representing a 74% recruitment rate. There were no significant differences between the included and excluded hospitals in teaching status, urban or rural location, or size. Discharges from the responding hospitals were weighted to account for nonresponse and to be representative of all discharges from psychiatric units at Pennsylvania acute care general hospitals.

We conducted a retrospective cross-sectional chart review with two stages: nurses trained as medical record administrators (MRAs) screened and flagged the random sample of medical records within each hospital for potential patient safety events, and trained physician reviewers then evaluated flagged charts for harm, error, and preventability. We developed training manuals and instrumentation to implement both tiers of the study on the basis of the methodology used in the landmark patient safety studies (2,19). We then recruited 11 MRAs across the state to review medical records from each study hospital and flag them as described above. After extensive training, the teams of MRAs reviewed 4,371 of the 4,401 medical records (99.3%); several charts were unavailable or unable to be located. Each medical record was carefully evaluated by using information in admission and discharge notes, clinical notes, nursing notes, progress notes, physician orders, and medication administration records. We recruited and implemented our rigorous training process with 12 physicians across the state. Throughout the study, reviewers examined an overlap of 10% of the sampled charts, and these were used as training files during weekly calls to maintain reliability. Our study team of reviewers abstracted data from the medical records of all hospitals in the sample, with the exception of one hospital that stipulated that only its employees could have access to the data. For that hospital, we trained a separate team of nurse and physician reviewers in the study methods. None of the reviewers had treated the patients whose records they reviewed.

Patient safety events were categorized as dichotomous outcomes for the occurrence of AEs or MEs. The following events were categorized as AEs: adverse drug event, self-harm or injury, assault, sexual contact, patient fall, and other. MEs included any mention in the chart of medication errors, elopement, possession of contraband, and other nonmedication errors. [Detailed definitions of these events are included in an online supplement to this article, and the events have been described elsewhere (17)].

For each discharge, we examined patient demographic and clinical factors, as well as characteristics of the hospital from which the patient was discharged. We investigated data on gender, age (18–30, 31–42, 43–53, and ≥54) and race (white or nonwhite). Clinical factors included admission day (weekday versus weekend), length of stay (one to three, four to six, seven to nine, and 10 or more days), insurance status (uninsured, commercial, or Medicaid or Medicare). The ICD-9 coding system was used to categorize principal diagnosis, which included psychosis (codes 295, 297, and 298), mood disorders (code 296), and other diagnoses (all other codes). Suicidal ideation (V62.84) or suicide attempts (E950–E959) were identified by using all available diagnosis codes on the discharge claim. Drug use was categorized as continuous (codes 303.01, 303.91 304.21–304.91, and 305.01), unspecified or episodic (303.00, 303.90, 304.20–304.90, and 305.20 or 303.02, 303.92, 304.22–304.92, and 305.02), or none or in remission (303.03, 303.93, 304.23–304.93, and 305.03). Hospital-level factors included teaching status (teaching or nonteaching); urbanity (urban or rural); and low, medium, high, or very high hospital volume based on annual admissions (0–800, 801–1,060, 1,061–1,280, and ≥1,281, respectively). Stratification of the continuous variables (age, length of stay in days, and number of admissions per year) was based on the median and interquartile range.

First, we used chi-square tests to compare demographic, clinical, and hospital factors between patients with and without an AE or ME. Second, we calculated the proportion of hospitalizations with an AE or ME, both overall and separately by each type of event. We then conducted unadjusted and adjusted logistic regression analyses to assess the strength of the relationship between the patient and hospital factors and the AE or ME. Odds ratios were adjusted for all demographic, clinical, and hospital factors, and we considered predictors in the adjusted analyses statistically significant at a significance level of <.05 (two-tailed). Finally, we present box-and-whisker plots to describe the distribution of unadjusted hospital AE and ME rates calculated by using Stata, version 14.0. Analyses were performed by using complex samples analysis in SPSS, version 24, which accommodated the study design’s two-stage proportional sampling and its weighting.

As with every study, this study had limitations. First, chart review studies may not document the full nature and outcomes of care. However, existing research shows that systematic methods of detecting AEs are 10 times more effective than other detection methods, such as voluntary reporting (27). Second, the use of administrative data to test patient and hospital predictors may have been influenced by the presence of measurement error for clinical characteristics (for example, diagnosis) that are not collected by validated instruments. Third, we did not have access to data about important covariates, such as comorbid medical issues that may complicate risk of AEs and MEs and socioeconomic status. Fourth, because only one hospital included in this study was a for-profit hospital, we were unable to examine the potential role of ownership in the occurrence of AEs and MEs. Fifth, the presence and extent of patient psychological harm were difficult to ascertain in a chart review; a trauma-informed lens that takes these factors into consideration should be employed when designing future research and safety interventions in inpatient psychiatry. Sixth, this study of inpatient psychiatric hospitalizations was conducted at general acute care hospitals and may not be applicable to psychiatric hospitals or longer-term units. For example, patients admitted to general hospitals are likely to have a shorter stay compared with patients admitted to stand-alone psychiatric hospitals. Finally, these findings may not be generalizable to other states given the role of varying state regulations (licensure, policies, insurance mandates, and so forth).

Essential to patient safety is understanding the underlying patterns of patient and provider factors that are associated with AEs and MEs. Ours is the first large-scale study that used an established methodology (17) to examine predictors of AEs and MEs in inpatient psychiatric units, providing new information about the patient and hospital factors associated with patient safety events among psychiatric inpatients of acute general care hospitals. Extensive interhospital variability exists in rates of AEs and MEs in medical-surgical hospitalizations (20,28), which suggests that “poor-performing” hospitals could learn from “high performers” by comparing hospitals systems, identifying gaps, and improving care. In general medicine, research has established a framework to lower rates of preventable harm (14) in five steps: measurement, evidence-based care practices, investment in implementation sciences, local ownership and peer learning, and aligning and synergizing efforts around common goals and measures. The field of mental health care would do well to adopt a similar framework in which this and other studies measuring AEs and MEs can serve as a first step. From our findings, we can then move to the next step and develop evidence-based practices that address the specific vulnerabilities to patient safety in inpatient psychiatry, by using high-quality interventions focused on improving care paths, such as “plan-do-study-act” cycles (29). These interventions should be targeted to patient groups with the highest risk of experiencing a patient safety event. The effectiveness of improvement interventions could be evaluated by using the recently published road map from an international consensus group in the field of psychiatry (30). Future studies should continue the steps along the framework in order to develop comprehensive safety improvements for this vulnerable patient population.