Clinical Trial Details
— Status: Recruiting
Administrative data
| NCT number |
NCT03626766 |
| Other study ID # |
AAAR0080 |
| Secondary ID |
|
| Status |
Recruiting |
| Phase |
N/A
|
| First received |
|
| Last updated |
|
| Start date |
September 1, 2018 |
| Est. completion date |
December 31, 2022 |
Study information
| Verified date |
August 2021 |
| Source |
Columbia University |
| Contact |
Jason Adelman, MD,MS |
| Phone |
646-317-4803 |
| Email |
adelman.jason[@]columbia.edu |
| Is FDA regulated |
No |
| Health authority |
|
| Study type |
Interventional
|
Clinical Trial Summary
This is a multi-site, cluster-randomized controlled trial to test the effectiveness of
patient photographs displayed in electronic health record (EHR) systems to prevent
wrong-patient order errors. The study will be conducted at three academic medical centers
that utilize two different EHR systems. Because EHR systems have different functionality for
displaying patient photographs, two different study designs will be employed. In Allscripts
EHR, a 2-arm randomized trial will be conducted in which providers are randomized to view
order verification alerts with versus without patient photographs when placing electronic
orders. In Epic EHR, a 2x2 factorial trial will be conducted in which providers are
randomized to one of four conditions: 1) no photograph; 2) photograph displayed in the banner
only; 3) photograph displayed in a verification alert only; or 4) photograph displayed in the
banner and verification alert. The main hypothesis of this study is that displaying patient
photographs in the EHR will significantly reduce the frequency of wrong-patient order errors,
providing health systems with the evidence needed to adopt this safety practice.
We will use the Wrong-Patient Retract-and-Reorder (RAR) measure, a valid, reliable, and
automated method for identifying wrong-patient orders, as the primary outcome measure. The
RAR measure identifies orders placed for a patient that are retracted within 10 minutes, and
then reordered by the same provider for a different patient within the next 10 minutes. These
are near-miss errors, self-caught by the provider before they reach the patient and cause
harm. In one study, the RAR measure identified more than 5,000 wrong-patient orders in 1
year, with a rate of 58 wrong-patient errors per 100,000 orders. Real-time telephone
interviews with clinicians determined that the RAR measure correctly identified near-miss
errors in 76.2% of cases. Thus, the RAR measure provides sufficient valid and reliable
outcome data for this study.
Description:
TThe main hypothesis of this study is that displaying patient photographs in electronic
health record (EHR) systems will significantly reduce the frequency of wrong-patient
electronic order errors.
SPECIFIC AIMS
Investigators will test this hypothesis using a multi-site randomized controlled trial and
pursue the following specific aims:
Aim 1: Test the effectiveness of displaying patient photographs in EHR systems for preventing
wrong-patient orders, using the Retract-and-Reorder measure to identify the outcome.
Aim 2: Identify characteristics of providers, patients, and orders that impact the
effectiveness of patient photographs displayed in an EHR system to prevent wrong-patient
orders.
METHODS Study Sites. The study will be conducted at NewYork-Presbyterian Hospital, Johns
Hopkins Medicine, and Montefiore Medical Center in two different electronic health record
(EHR) systems. Five NewYork-Presbyterian Hospital campuses that utilize Allscripts EHR will
be included: Milstein Hospital, Weill Cornell Medical Center, Morgan Stanley Children's
Hospital of New York, Lower Manhattan Hospital, and the Allen Hospital. In addition,
affiliated Ambulatory Care Network outpatient sites will be included. The effectiveness of
patient photographs in Epic will be tested at Johns Hopkins Hospital and three Montefiore
Medical Center campuses (Moses Hospital, Weiler Hospital, Wakefield Hospital).
Research Design. The main hypothesis of this study is that displaying patient photographs in
an EHR system at the time of placing electronic orders will significantly decrease the
frequency of wrong-patient orders. Investigators will test this hypothesis with a randomized
controlled trial using a 2-arm design in Allscripts and a 2 x 2 factorial design in Epic to
determine the optimal configuration of patient photographs. The design at the different sites
is based on the functionality of the EHR systems. In Allscripts, providers randomized to the
intervention arm will be shown a verification alert with a patient photograph at the time of
placing orders; providers randomized to the control arm will be shown the verification alert
with an avatar indicating the patient's sex. Epic has the functionality to display a patient
photograph in the banner at the top of the screen and in a verification alert at the time of
placing orders. Therefore, in Epic at Johns Hopkins and Montefiore, providers will be
randomized to one of four conditions: no photograph; photograph displayed in the banner only;
photograph displayed in the verification alert only; or photograph displayed in the banner
and alert.
Patient Inclusion Criteria. In this study, all patients for whom an order is placed during
the study period at the study sites will be included in the analysis. At
NewYork-Presbyterian, all patients age 5 years and older will be included; at Johns Hopkins
Hospital and Montefiore Medical Center, all patients age 2 years and older will be included.
Obtaining a photograph at registration is standard procedure at the participating study
sites. Patients will not be required to provide written or oral consent to have their
photograph taken, as the photographs are used as part of routine care. However, patients or
their legal guardians may refuse.
Provider Inclusion Criteria. Any provider who places an electronic order can potentially
place an order on the wrong patient. Therefore all randomized providers who place an
electronic order at the study sites during the course of the study period will be included.
Randomization. In Allscripts at NewYork-Presbyterian, the EHR system has been configured such
that the first time a provider opens the Order Entry screen s/he will be automatically
randomized to either the control or intervention group (1:1 per the 2-arm design) using a
computerized randomization algorithm. In Epic, randomization will be performed using the same
randomization algorithm; however, providers will be manually assigned to the control or
intervention groups (1:1:1:1 per the 2 x 2 design) using the Grouper function in Epic.
Primary Outcome. The primary outcome is the frequency of Wrong-Patient Retract-and-Reorder
(RAR) events, identified using the RAR measure and defined as an order that is placed for a
patient, retracted within 10 minutes of placing the order, and then reordered by the same
provider for a different patient within 10 minutes of the retraction.
Clustering of Orders within Order Sessions. If a provider begins placing orders in the wrong
patient's record, there is the possibility that several such orders will be placed
consecutively and then all retracted together. Therefore, individual orders do not represent
independent opportunities for RAR events to occur. Rather, orders are clustered within order
sessions. An order session is defined as a series of orders placed by a provider on a single
patient that begins with opening that patient's order file and terminates when an order is
placed on another patient or after 60 minutes, whichever comes first.
Unit of Analysis. The unit of analysis will be the order session. Provider Level, Patient
Level, Order-Session Level, and Order Level Covariates. The data for this study will have a
nested, hierarchical structure with orders clustered within order sessions, and order
sessions clustered within providers. The analysis will account for this hierarchical
structure. Additional variables will be extracted from the electronic medical record,
including attributes of the provider (attending, resident, physician assistant, nurse
practitioner, or other), patient (age, race, ethnicity, sex, insurance status, unit), order
session (location, duration, number of orders), and order (medication, radiology, lab,
other). The presence or absence of a patient photograph will be tracked as an order level
covariate.
Primary Analysis. The unit of analysis is the order session. The dependent variable is the
proportion of order sessions that contain at least one order that was retracted and
re-ordered (an RAR event). Inference about effectiveness of the intervention will be based on
a Wald test of the coefficient of an interaction term between study-arm assignment and an
indicator of pre- or post-intervention time period. The estimate of effectiveness will be the
odds ratio (exponentiated coefficient) reported with its 95% confidence interval. With
cluster randomization at the provider level, patient and order session level attributes may
not be evenly distributed across groups. To reduce confounding bias, covariates at these
levels that are both differentially distributed in the study groups and associated with the
outcome will be included in this primary analysis. Provider-level covariates will likely be
balanced across study groups by randomization, and will be included in this primary analysis
only if descriptive statistics show otherwise.
2-Arm RCT: Statistical Power/Sample Size. The overall design is a cluster randomized trial,
with order-session observations nested within randomized providers. The number of providers
(clusters) is a fixed attribute of the study site and will not be subject to investigator
control. Based on preliminary data, the assumption is that data on 12,000 providers will be
collected, with half randomized to each study arm. Over 2.5 years of observation, it is
estimated that an average of 5,000 order sessions per provider will be accrued, with a
coefficient of variation of 1.27. An intra-provider correlation of 0.001 for the outcome is
anticipated. The most recent available analyses from NYP suggest that the wrong-patient order
session rate (a weighted average of inpatient and outpatient rates) is about 130 per 100,000
order sessions. Using a two-tailed test at the 0.05 significance level, this will provide
>90% power to detect a 25% reduction in the wrong-patient order-session rate in the
intervention group in the primary analysis.
2x2 Factorial RCT: Statistical Power/Sample Size. A statistical power simulation was
conducted for the 2x2 factorial study of banner photographs (photo/no photo in banner) and
alert photographs (alert with photo/no alert). Based on prior studies and guidance from study
investigators, it is estimated that approximately 8,268 providers will be randomized in a
1:1:1:1 ratio to the four study arms (no photograph, photograph in the banner only,
photograph in the verification alert only, or photograph in the banner and verification
alert). Providers were assumed to generate an average of 74 order sessions per month, with a
Poisson distribution, and that data collection would continue for 1 year. Based on previous
studies, the variance component (in the log-odds metric) at the provider level would be 1.9
(corresponding to an intra-provider correlation of 0.52). The base rate of RAR events was
assumed to be 90 per 100,000 order sessions. Based on 400 simulations conducted under these
assumptions, it was determined that the study would have 91% power (95% CI 88%-94%) to detect
an odds ratio of 0.75 (i.e., 25% reduction in RAR events) for either single-photograph
intervention compared to no intervention. The power to detect an additional 15% reduction by
the combination of both photographs (compared to either alone) is 99.8% (95% CI 98.6%-99.9%).
Missing Data. Due to the automatic functioning of the EHR, no missing data concerning
Retract-and-Reorder events, the mode of intervention group versus control group, or the
provider-level covariates is expected. There may be sporadic missing information regarding
the patient-level covariates. If any variables are missing for more than 1 record per 1,000
(after backfilling based on other records involving the same patient), analyses will be
extended to address this. The data will be presumed to be missing at random and will apply
multiple imputation with chained equations.
Interim Analysis. To safeguard against the possibility that the intervention actually worsens
(increases) the rate of RAR events, and to prevent unnecessary continuation of a study that
is already conclusive, a data safety monitoring committee will conduct one interim review of
the data. At the midpoint of the study, 50% of the data is expected to have been accrued.
Using the Lan-Demets alpha spending procedure with symmetric O'Brien-Fleming boundaries, the
stopping rule at the interim review will be a z-statistic of magnitude 3.0318 or greater. The
associated nominal P value is .0024. Combined with a final analysis using a critical z-value
of 1.9669 (nominal P = .0492), the overall alpha of 0.05 will be spent by the end of the
study. The effects of this interim analysis procedure on nominal statistical power (see
above) is negligible, less than 0.5 percentage points, so no adjustments to data collection
are needed to account for this.
Intention-to-Treat vs As-Treated Analyses. Some patients will likely decline to be
photographed or for whom photographs are not taken for various reasons. The photo capture
rate (the number of patients with a photograph divided by the number of patients with a visit
during the study period) will be monitored throughout the study period. In the primary
analyses for all trials, an intention-to-treat principle will be used, and these patients'
and sites' records will be included in the analysis as if they had been photographed and
participated. A second "as-treated" analysis will be performed limited to compliant patients
and sites, recognizing that such analyses may be subject to confounding by patient and site
factors.
