Pictographs for Preventing Wrong-Patient Errors in NICUs

BACKGROUND Newborns in the neonatal intensive care unit (NICU) are at high risk for wrong-patient errors. Effective 2019, The Joint Commission requires that health systems adopt distinct methods of newborn identification to prevent this type of error as part of its National Patient Safety Goals. Displaying patient photographs in the electronic health record (EHR) is a promising strategy to improve identification of children and adults, but photographs are not likely to be effective for identifying newborns. Therefore, this study assesses the use of Pictographs as a "photo equivalent" for improving identification of newborns in the NICU. METHODS The primary hypothesis of this study is that clinicians randomly assigned to view Pictographs in the EHR (intervention arm) will have a significantly lower rate of wrong-patient order errors in the neonatal intensive care unit (NICU) compared to clinicians assigned to view no Pictographs in the EHR (control arm). The secondary hypothesis is that the rate of wrong-patient orders for siblings of multiple births in the NICU will be significantly lower for clinicians randomized to the intervention versus control arm. The study will pursue the following specific aims: Aim 1: Test the effectiveness of displaying Pictographs in EHR systems for reducing wrong-patient orders among infants in the NICU, using the Wrong-Patient Retract-and-Reorder measure to identify the outcome. Aim 2: Conduct subgroup analyses of the effectiveness of Pictographs for reducing the rates of wrong-patient orders among siblings of multiple-birth infants in the NICU. Design. This multi-site, two-arm, parallel-group, cluster randomized controlled trial will test the effectiveness of Pictographs for preventing wrong-patient order errors in the NICUs of three health systems. Pictographs consist of three elements: 1) pictorial symbols of easy-to-remember objects (e.g., rainbow, lion); 2) the infant's given name (when available); and 3) a color-coded border indicating the infant's sex. For each infant admitted to the study NICUs, parents or guardians will be asked to select a unique Pictograph to be displayed on the isolette and in the EHR for the duration of each infant's hospital stay. Clinicians randomized to the intervention arm will see a Pictograph displayed in the banner and verification alert in the EHR. Clinicians randomized to the control arm will not see Pictographs in the EHR. Study Sites. The study includes NICUs at the following participating study institutions representing diverse patient populations: NewYork-Presbyterian Hospital, New York, NY; Montefiore Medical Center, Bronx, NY; and Johns Hopkins Medicine, Baltimore, MD. All sites will conduct the study in EpicCare (Epic Systems Corporation, Verona, WI). Patient Inclusion Criteria. All infants admitted to the study NICUs will be eligible to be included. Clinician Inclusion Criteria. All clinicians who place electronic orders in the NICU at the study sites during the study period will be included. Randomization. The clinician is the unit of randomization. Each ordering clinician will be randomly assigned to a study arm in a 1:1 random allocation scheme using a computer random number generator program at all study sites. Then, clinicians will be assigned, as randomized, to groupers built into Epic to apply configuration rules that dictate whether Pictographs are displayed according to study arm. Primary Outcome. The primary outcome is wrong-patient order sessions, defined as a series of orders placed for a single patient by a single clinician that contains at least one wrong-patient order. The Wrong-Patient Retract-and-Reorder (RAR) measure--a validated, reliable, and automated method endorsed by the National Quality Forum (NQF #2723)--will be used to identify wrong-patient orders. The Wrong-Patient RAR measure identifies RAR events, defined as one or more orders placed for a patient that are retracted within 10 minutes, and then reordered by the same clinician for a different patient within the next 10 minutes. These are near-miss errors, caught by the clinician before they reach the patient. In the validation study, real-time telephone interviews with clinicians demonstrated that 76.2% of RAR events identified by the measure were confirmed to be wrong-patient orders. In a study conducted at a large academic medical center, the RAR measure identified more than 5,000 wrong-patient orders in 1 year. Thus, the RAR measure provides robust outcome data to test safety interventions. Unit of Analysis. If a clinician begins placing orders in the wrong patient's record, several such orders may be placed consecutively and then retracted together. Therefore, individual orders do not represent independent opportunities for RAR events to occur, rather the order session represents an independent opportunity for an RAR event to occur. Thus, the unit of analysis is the order session, defined as a series of orders placed by a clinician for a single patient that begins with opening that patient's order file and terminates when an order is placed for another patient or after 60 minutes, whichever occurs first. The presence of Pictographs will be tracked as an order-session level characteristic. Higher Level Units of Analysis. Order sessions are nested within both patient admission and clinician, and both patients and clinicians are nested within NICU. In the analysis, NICU will be treated as a fixed covariate, and clinicians are treated as an important source of variance and "cluster" unit of analysis because clinician is the unit of randomization. Data Extraction and Management. Data will be extracted retrospectively for all orders (e.g., medications, laboratory tests, imaging studies, and other types) placed by randomized clinicians for patients receiving care in the study NICUs during the study period. Data for each order, including order, patient, and clinician characteristics, will be extracted from the health system data warehouse at each study site at the end of the study period. All patient and clinician identifiers will be replaced by pseudo-identifiers by research personnel at study sites to create de-identified analytic data sets, and data will be transferred electronically to the study biostatistician using a secure file transfer protocol. Missing Data. Due to the automatic functioning of the EHR, no missing data concerning orders, RAR events, study group assignment, or provider-level covariates is expected. There may be missing information regarding patient-level covariates. If any variables are missing for more than 1 record per 1,000, analyses will be extended to address this. Data will be presumed to be missing at random and multiple imputation with chained equations will be applied. Statistical Analysis Plan. Because order sessions are nested within clinicians, Aim 1 will be tested by estimating a multilevel mixed Poisson regression model predicting RAR event status (yes, no) with Group (Pictograph, coded 1, vs no Pictograph, coded 0) as the primary predictor and Multiple Status and NICU included as categorical covariates ("class" variables in SAS); the intercept will be treated as a random effect, assumed to vary among clinicians within each group. Because RAR events are rare (approximately 5 per 10,000 order sessions), the mixed Poisson regression model and the mixed logistic regression model yield nearly identical results with parameter estimates differing in the 3rd significant digit. The hypothesis will be confirmed if the RR for Group is significantly different from 1.0 (expected to be \<1.0), based on the Wald chi-square test of BGroup; i.e., ln(RR). To test Aim 2, a variant of the same multilevel Poisson regression model will be estimated in which the main effect of Group is replaced with the Multiple Status\*Group interaction; this specification will yield two coefficients for Group, one for Singletons (Multiple Status=0) and one for Multiples (Multiple Status=1). The Wald chi-square test will be used to evaluate the statistical significance of each coefficient, i.e., the effect of Pictographs on RAR risk for singletons and for multiples. The SAS GLIMMIX procedure will be used to perform these analyses, and a 2-tailed, alpha=0.05 criterion will be used to evaluate statistical significance. Secondary analyses will examine whether the effect of the Pictograph intervention varies across hospitals, by EHR vendor, by clinician type and other clinician characteristics, by patient characteristics, and/or by order session characteristics. Statistical Power/Sample Size. The power was calculated using the O'Brien and Muller approach, an exemplary data set constructed by assumed population parameters to detect the hypothesized effect. It is estimated approximately 17,000 patients, infants admitted to the NICU at each site will be recruited, and verbal consent will be obtained from parents or legal guardians. The number of clinicians is a fixed attribute of the study sites and is not subject to investigator control. Based on preliminary data, it is assumed that data will be obtained on 6,250 clinicians, with half randomized to each study arm. With respect to Aim 1, if the effect sizes are reductions of 30% for singletons and 50% for multiples, as hypothesized, the study will have 98% power to detect a significant benefit of Pictographs on RAR event rate. By changing the weights described above to 0.75 and 0.25 for both singletons and multiples, it is estimated that the study will have 81% power to detect an overall reduction, across singletons and multiples, of 25% in RAR event rate. Using the mean standard error, the calculated power for Aim 2 of the proposed study: for singletons, the power to detect the hypothesized 30% reduction in RAR rate is 89%; for multiples, the power to detect a 50% reduction is 95%. Minimum effect sizes detectable with 80% power are reductions of 27% for singletons and 42% for multiples. Interim Analysis. At the midpoint of the study, the Lan-Demets algorithm will be applied for early termination with an O'Brien-Fleming boundary function to preserve overall Type I error rate. At the interim review, the corresponding stopping points correspond to a z-statistic for the primary hypothesis test of ±2.9626. To avoid bias in these judgments, the data sets will contain interim results for the two arms but will not identify which arm is the intervention. A test of futility will be conducted, by treating the expected result of 25% reduction as a null hypothesis. If that hypothesis test is rejected at the 1-tailed 0.001 level, continuation will be considered futile and will conclude the study. Recruitment/Enrollment. A nurse involved in patient care will inform parents or guardians of infants admitted to the NICU about the study. The nurse will obtain verbal consent to display the Pictograph on the infant's isolette and in the infant's medical record in the EHR for the duration of the infant's hospital stay. Parents who provide verbal consent will select a unique Pictograph for each infant using the Pictograph Application on an electronic tablet. The Pictograph Application is a HIPAA-compliant, secure web-based application that enables Pictograph selection, customization, printing, and uploading to the EHR. Parents with multiple infants in the NICU will select a different Pictograph for each infant. Selected Pictographs become unavailable such that no two infants will have the same Pictograph at the same time in the NICU. The selected Pictograph will be printed and attached to the isolette, and uploaded into the EHR using the same functionality used to upload and display patient photographs. When the infant is discharged, the Pictograph will be removed from the isolette and from the EHR; that image will remain unavailable for use for 2 weeks after the infant is discharged.