Zahir Kanjee

Zahir Kanjee

Auto-generated profile, based on publicly available information

Zahir Kanjee

Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts (H.L., Z.K.)

Research Focus & Metrics

Academic Career Length

16Years (2011-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
5 February 2025 in Nature Medicine
Verified
Journal Article
Closed access
PDF file
EGEthan Goh
Ethan Goh
RGRobert J. Gallo
Robert J. Gallo
ESEric Strong
Eric Strong
YWYingjie Weng
Yingjie Weng
HKHannah Kerman
Hannah Kerman
ARAdam Rodman
Adam Rodman

30 April 2026 in Science
Verified
Journal Article
Closed access
PBPeter G. Brodeur
Peter G. Brodeur
TBThomas A. Buckley
Thomas A. Buckley
ZKZahir Kanjee
Zahir Kanjee
EGEthan Goh
Ethan Goh
ELEvelyn Bin Ling
Evelyn Bin Ling
ARAdam Rodman
Adam Rodman
More than 65 years ago, complex clinical diagnostic reasoning cases were introduced as the gold standard for the evaluation of expert medical computing systems, a standard that has held ever since. In this study, we report the results of a physician evaluation of a large language model (LLM) on challenging clinical cases across five experiments with a baseline of hundreds of physicians. We then report a real-world study comparing human expert and artificial intelligence (AI) second opinions in randomly selected patients in the emergency room of a major tertiary academic medical center. In all experiments, the LLM outperformed physician baselines and displayed continued improvement from prior generations of AI clinical decision support. Our study suggests that LLMs have eclipsed most benchmarks of clinical reasoning, motivating the urgent need for prospective trials.

1 December 2011 in Journal of Hospital Infection
Verified
Journal Article
Closed access
ZKZ. Kanjee
Z. Kanjee
KCK. Catterick
K. Catterick
AMA.P. Moll
A.P. Moll
KAK.R. Amico
K.R. Amico
GFG.H. Friedland
G.H. Friedland

2 August 2016 in PLOS Medicine
Verified
Journal Article
Open access
PDF file
JLJohn Ly
John Ly
VSVidiya Sathananthan
Vidiya Sathananthan
TGThomas Griffiths
Thomas Griffiths
ZKZahir Kanjee
Zahir Kanjee
AKAvi Kenny
Avi Kenny
JKJohn D. Kraemer
John D. Kraemer
The Ebola virus disease (EVD) epidemic has threatened access to basic health services through facility closures, resource diversion, and decreased demand due to community fear and distrust. While modeling studies have attempted to estimate the impact of these disruptions, no studies have yet utilized population-based survey data. We conducted a two-stage, cluster-sample household survey in Rivercess County, Liberia, in March–April 2015, which included a maternal and reproductive health module. We constructed a retrospective cohort of births beginning 4 y before the first day of survey administration (beginning March 24, 2011). We then fit logistic regression models to estimate associations between our primary outcome, facility-based delivery (FBD), and time period, defined as the pre-EVD period (March 24, 2011–June 14, 2014) or EVD period (June 15, 2014–April 13, 2015). We fit both univariable and multivariable models, adjusted for known predictors of facility delivery, accounting for clustering using linearized standard errors. To strengthen causal inference, we also conducted stratified analyses to assess changes in FBD by whether respondents believed that health facility attendance was an EVD risk factor. A total of 1,298 women from 941 households completed the survey. Median age at the time of survey was 29 y, and over 80% had a primary education or less. There were 686 births reported in the pre-EVD period and 212 in the EVD period. The unadjusted odds ratio of facility-based delivery in the EVD period was 0.66 (95% confidence interval [CI] 0.48–0.90, p-value = 0.010). Adjustment for potential confounders did not change the observed association, either in the principal model (adjusted odds ratio [AOR] = 0.70, 95%CI 0.50–0.98, p = 0.037) or a fully adjusted model (AOR = 0.69, 95%CI 0.50–0.97, p = 0.033). The association was robust in sensitivity analyses. The reduction in FBD during the EVD period was observed among those reporting a belief that health facilities are or may be a source of Ebola transmission (AOR = 0.59, 95%CI 0.36–0.97, p = 0.038), but not those without such a belief (AOR = 0.90, 95%CI 0.59–1.37, p = 0.612). Limitations include the possibility of FBD secular trends coincident with the EVD period, recall errors, and social desirability bias. We detected a 30% decreased odds of FBD after the start of EVD in a rural Liberian county with relatively few cases. Because health facilities never closed in Rivercess County, this estimate may under-approximate the effect seen in the most heavily affected areas. These are the first population-based survey data to show collateral disruptions to facility-based delivery caused by the West African EVD epidemic, and they reinforce the need to consider the full spectrum of implications caused by public health emergencies.

1 February 2012 in Journal of Infection and Public Health
Verified
Journal Article
Closed access
ZKZ. Kanjee
Z. Kanjee
KAK.R. Amico
K.R. Amico
FLF. Li
F. Li
KMK. Mbolekwa
K. Mbolekwa
AMA.P. Moll
A.P. Moll
GFG.H. Friedland
G.H. Friedland

14 March 2024 in medRxiv
Verified
Preprint
Open access
PDF file
EGEthan Goh
Ethan Goh
RGRobert Gallo
Robert Gallo
JHJason Hom
Jason Hom
ESEric Strong
Eric Strong
YWYingjie Weng
Yingjie Weng
JCJonathan H Chen
Jonathan H Chen
Importance Diagnostic errors are common and cause significant morbidity. Large language models (LLMs) have shown promise in their performance on both multiple-choice and open-ended medical reasoning examinations, but it remains unknown whether the use of such tools improves diagnostic reasoning. Objective To assess the impact of the GPT-4 LLM on physicians’ diagnostic reasoning compared to conventional resources. Design Multi-center, randomized clinical vignette study. Setting The study was conducted using remote video conferencing with physicians across the country and in-person participation across multiple academic medical institutions. Participants Resident and attending physicians with training in family medicine, internal medicine, or emergency medicine. Intervention(s) Participants were randomized to access GPT-4 in addition to conventional diagnostic resources or to just conventional resources. They were allocated 60 minutes to review up to six clinical vignettes adapted from established diagnostic reasoning exams. Main Outcome(s) and Measure(s) The primary outcome was diagnostic performance based on differential diagnosis accuracy, appropriateness of supporting and opposing factors, and next diagnostic evaluation steps. Secondary outcomes included time spent per case and final diagnosis. Results 50 physicians (26 attendings, 24 residents) participated, with an average of 5.2 cases completed per participant. The median diagnostic reasoning score per case was 76.3 percent (IQR 65.8 to 86.8) for the GPT-4 group and 73.7 percent (IQR 63.2 to 84.2) for the conventional resources group, with an adjusted difference of 1.6 percentage points (95% CI -4.4 to 7.6; p=0.60). The median time spent on cases for the GPT-4 group was 519 seconds (IQR 371 to 668 seconds), compared to 565 seconds (IQR 456 to 788 seconds) for the conventional resources group, with a time difference of -82 seconds (95% CI -195 to 31; p=0.20). GPT-4 alone scored 15.5 percentage points (95% CI 1.5 to 29, p=0.03) higher than the conventional resources group. Conclusions and Relevance In a clinical vignette-based study, the availability of GPT-4 to physicians as a diagnostic aid did not significantly improve clinical reasoning compared to conventional resources, although it may improve components of clinical reasoning such as efficiency. GPT-4 alone demonstrated higher performance than both physician groups, suggesting opportunities for further improvement in physician-AI collaboration in clinical practice.