Stephenson is a doctoral candidate in public policy at Cornell University. He has research interests in health economics, organizational economics, and physician behaviour. He is a trained family physician and practices as an emergency physician and as a family physician in remote and northern communities. He will be available for interviews for the 2022-2023 job market.
PhD in Public Policy, 2023 (Expected)
Brooks School of Public Policy, Cornell University
Residency in Family Medicine, 2017
University of Toronto
Medical doctorate, 2015
University of Manitoba
MA in Economics, 2011
BA in Economics, 2010
We assessed whether the timing and order of patients over emergency shifts are associated with receiving diagnostic imaging in the emergency department and characterized whether changes in imaging are associated with changes in patients returning to the ED. In this retrospective study, we used multivariate and instrumental variable regressions to examine how the timing and order of patients are associated with the use of diagnostic imaging. Outcomes include whether a patient receives a radiograph, a computed tomography (CT) scan, an ultrasound, and 7-day bouncebacks to the ED. The variables of interest are time and order during a physician’s shift in which a patient is seen.A total of 841,683 ED visits were examined from an administrative database of all ED visits to Niagara Health. Relative to the first patient, the probability of receiving a radiograph, CT, and ultrasound decreases by 6.4%, 9.1%, and 3.8% if a patient is the 15th patient seen during a shift. Relative to the first minute, the probability of receiving a radiograph, CT, or ultrasound increases by 1.9%, 2.7%, and 1.1% if a patient is seen in the 180th minute. Seven-day bounceback rates are not consistently associated with patient order or timing in a shift and imaging orders. Imaging in the ED is associated with shift length and especially patient order, suggesting that physicians make different imaging decisions over the course of their shifts. Additional imaging does not translate into reductions in subsequent bouncebacks to the hospital.
The purpose of this research is to examine the prevalence and characteristics of influenza-like illness (ILI) related presentations among people experiencing homelessness compared to the general population as well as to use the Susceptible, Infected, Recovered (SIR) simulation model parameters β and γ to model infectious interactivity, recovery rate, and population-level basic reproduction number (R0). Using administrative health data from emergency department (ED) visits in the province of Ontario, Canada from 2010 to 2017, an SIR model was used to calculate the R0 for ILI in both the general population and the population of homeless individuals. From 2010 to 2017, a total of 17,056 homeless and 85,553 non-homeless individuals presented with an ILI to an ED in Ontario. The estimated infectious interactivity (β) was lower while the recovery rate (γ) was longer for infected people experiencing homelessness. Our results suggest that infections of ILI will result in more secondary cases in the homeless population compared to the homed population. This evaluation of the dynamics of ILI spread in the homeless population provides insight into how illnesses such as COVID-19 may be much more infectious in this population compared to the homed population.
The COVID-19 pandemic has placed unprecedented strain on healthcare systems and may have consequential impacts on patient safety incidents (PSIs). The primary objective of this study was to examine the impact of the COVID-19 pandemic on PSIs reported in Niagara Health. Flexible Farrington models were used to retrospectively detect weeks from January to September 2020 where PSI counts were significantly above expected counts. Incident counts were adjusted to weekly inpatient-days. Outcomes included overall incident numbers, incidents by category, and incidents by ward type. The overall number of PSIs across Niagara Health did not increase during the first wave of the COVID-19 pandemic. However, significant increases in falls were observed, suggesting that other types of incidents decreased. Falls increased by 75% from February to March 2020, coinciding with the onset of the first wave of the pandemic. Further investigation by unit type revealed that the number of falls increased specifically on internal medicine and complex continuing care wards. Despite no observed changes in overall number, significant composition shifts in PSIs occurred during the first wave of the COVID-19 pandemic, with increased falls on internal medicine and complex continuing care wards. Possible explanations include restrictions on patient visitation, reduced patient contact/supervision, and/or personal protective equipment requirements. Providers should maintain a particularly high vigilance for patient falls during pandemic outbreaks, and hospitals should consider targeting resources to higher-risk locations. The results of this study reinforce the need for ongoing pandemic PSI monitoring and rapidly adaptive responses to new patient safety concerns.
The association between the nurse-to-patient ratio and patient outcomes has been extensively investigated. Real time location systems have the potential capability of measuring the actual amount of bedside contact patients receive. This study aimed to determine the feasibility and accuracy of real time location systems as a measure of the amount of contact time that nurses spent in the patients’ bed space. An exploratory, observational, feasibility study was designed to compare the accuracy of data collection between manual observation performed by a researcher and real time location systems data capture capability. Four nurses participated in the study, which took place in 2019 on two hospital wards. They were observed by a researcher while carrying out their work activities for a total of 230 minutes. The amount of time the nurses spent in the patients’ bed space was recorded in 10-minute blocks of time and the real time location systems data were extracted for the same nurse at the time of observation. Data were then analysed for the level of agreement between the observed and the real time location systems measured data, descriptively and graphically using a kernel density and a scatter plot. The difference (in minutes) between researcher observed and real time location systems measured data for the 23, 10-minute observation blocks ranged from zero (complete agreement) to 5 minutes. The mean difference between the researcher observed and real time location systems time in the patients’ bed space was one minute (10% of the time). On average, real time location systems measured time in the bed space was longer than the researcher observed time. There were good levels of agreement between researcher observation and real time location systems data of the time nurses spend at the bedside. This study confirms that it is feasible to use real time location systems as an accurate measure of the amount of time nurses spend at the patients’ bedside.
One proposed solution to prolonged emergency department (ED) wait times is a publicly available website that displays estimated ED wait times. This could provide information to patients so that they may choose sites with low wait times, which has the potential to smooth the overall wait times in EDs across a health system. We describe the effect of a novel city-wide ED wait time website on patient volume distributions throughout the city of Hamilton, Ontario, Canada. We compared the number of new patients arriving every 15 minutes during 2 separate time periods—before and after a publicly viewable wait time website was made available. For each ED site, the effect of the posted wait time was measured by assessing its association with the total number of patient arrivals in the subsequent hour at the same site and at all other sites in Hamilton. Linear models showed clinically modest changes in patient volumes when wait times changed. However, nonlinear models showed that a 60-minute increase in wait time at a site was associated with 10% fewer patients presenting over the next hour. Larger negative associations were observed at community hospitals and urgent care centers. Increases in wait times at a given site were also associated with increased patient volumes at other sites in the system. After the implementation of a public wait time website, elevated wait times led to fewer patients at the same site but more patient visits at other sites. This may be consistent with the wait time tracker inducing patients to avoid sites with high wait times and instead visit alternate sites in Hamilton, but only when wait times were very high.
There is currently interest in extending medical residency training in several countries including the United States and Canada. There is little evidence on what impacts extending medical residency has on supply of independent practitioners.I leverage the province-by-province roll-out of mandatory family medicine residencies in Canada from 1976 to 1994. This mandated that practitioners had to complete a two year residency instead of a one year internship. I use annual Canadian Institute of Health Information data on supply of physicians by specialty and province. I employ a difference-in-differences estimation strategy comparing specialities impacted by the legislation to those that had no change in their residency length (first difference). I compare before and after legislation by province (second difference). I find reductions in the supply of family medicine practitioners in the range of 3-5% of overall supply after implementation of a longer residency. This reduction is statistically significant lasting five years after mandate and point estimates of supply do not return to baseline until eight years after mandates. I find increases in the number of graduates of other programs that might plausibly substitute for family medicine suggestive that the policy drove medical students towards other residencies. Extending residency length has the potential to cause declines in physician supply over the short to medium run. There are both direct effects on physician supply through delays in cohorts as well as indirect effects through substitution away from family medicine residencies.
What are the impacts of medical segregation? In 1990, South Africa repealed legislation enacted during Apartheid to segregate medical care. This made it legal for Black Africans to use medical clinics, hospitals and wards that were of higher quality and previously exclusively reserved for whites. We use data from the South African census, USAID Demographic and Health Surveys, the Project for Statistics on Living Standards and Development and a database of hospital and clinic locations in South Africa in 1990. Using a birth cohort differences-in-differences regression we compare out- comes between Black South African children and other more advantaged South African children (first difference) born just before and just after desegregation (second difference). As an alternate differences-in-differences we examine Black South African children born just before and just after (first difference) this policy change to evaluate effects on child mortality through access to local white medical facilities. We compare areas that had a segregated white or mixed race clinic to places that did not and where service access would not have changed (second difference). We find a 5% decrease in the number of “missing” Black South African children per household and a 20% reduction in the mortality of the last black child born to a family after desegregation. Black children are 3% less likely to have a disability, 25% less likely to have a physical disability, and 10% less likely to have a seeing disability after access to better health care but this difference dissipates by ten years after the policy change. We find no evidence of improvement in anthropometric outcomes but we find increased contacts with the formal health care system and modestly improved vaccine uptake. Depending on the baseline infant mortality rate, we estimate that desegregation of the medical system prevented 1200 to 4800 deaths per year among Black children.
Doctors operate in high stakes environments where they need to make decisions quickly. This necessitates rapid collection and synthesis of information. Triage is one such mechanism that allows for speedy decisions. The goal of triage is to condense a vector of information into a single measure which is assigned to a patient to reflect the relative urgency of their case. Providers can then allocate emergency resources accordingly. However, triage represents a simplification of information that could obscure important details of a patients case. Does triage information change how providers allocate resources for patients? I examine a set of EDs that employ a five-point triage system. A patient’s score is based on arbitrary cut-offs of vital signs, such as the heart rate, which creates a regression discontinuity that quasi-randomly assigns a triage score. As a result of a worse triage score, I find that ED physicians increase the time they treat patients by 25% from baseline, and the amount that they order imaging tests by 10%. They reduce the probability of a patient receiving a procedure by 25%. They increase the probability of specialist consultation and admission to hospital. These changes in resource allocation suggest that the physician mitigates an increased signal of risk by collecting more information, treating less, and consulting colleagues more. There are no impacts on patient return to ED rates, suggesting increased resource allocation does not result in improved health.
TA: Spring 2022
TA: Spring 2022
TA: Fall 2021
TA: Spring 2021
TA: Fall 2020
TA: Fall 2018