Max Ward

Dr. Max Ward


Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

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

Representative Publications

View All Publications
20 June 2021 in British Journal Of Neurosurgery
Verified
Journal Article
Closed access
MWMax Ward
Max Ward
MAMohsen Ahmed
Mohsen Ahmed
CMChristopher Markosian
Christopher Markosian
JEJudith Z. Ezike
Judith Z. Ezike
RARishav Agrawal
Rishav Agrawal
AMAntonios Mammis
Antonios Mammis
Deep brain stimulation (DBS) is a common surgical option for the treatment of medically refractory Parkinson’s disease (PD). Manufacturer and User Facility Device Experience (MAUDE), a United State...

22 December 2025 in European Spine Journal
Verified
Journal Article
Open access
PDF file
AMApratim Maity
Apratim Maity
EBEthan D. L. Brown D.L. Brown
Ethan D. L. Brown D.L. Brown
MWMax Ward
Max Ward
EOEmily A. Orsino
Emily A. Orsino
DSDaniel Schneider
Daniel Schneider
SLSheng-fu Larry Lo
Sheng-fu Larry Lo
Purpose: This study aimed to analyze adverse events associated with Mazor X surgical guidance robots, focusing on parameters such as root causes, vertebral distribution, surgical delay, robot abandonment, and more. Methods: 330 adverse events reported within the Manufacturer and User Facility Device Experience (MAUDE) database for the Mazor X (MX) and Mazor X Stealth Edition (MXSE) robot guidance systems were analyzed and categorized based on the root causes, vertebral distribution, surgical delay, robot abandonment, and patient outcomes. Results: Both the MX and MXSE adverse events were primarily due to external root causes (35% and 52%, respectively), with user error being the most common. Deviations were most likely to affect the L3, L4, and L5 vertebra, and deviate either medially or laterally. Motor deficits were the most common adverse patient outcome for MX, while sensory deficits were the most common for MXSE. There were two deaths associated with the MX, compared to zero deaths with the MXSE. Conclusion: While the MAUDE database cannot be used to calculate the incidence or prevalence of adverse events, it provides insight into the types of adverse events that tend to occur with medical devices. MAUDE reports suggest that user-related errors account for most of the adverse events that occur, underscoring the need for focused surgeon training. Our study highlights the importance of understanding both the types of adverse events and their root causes to promote patient safety and reduce complication rates.

6 August 2025 in AI and Ethics
Verified
Journal Article
Open access
PDF file
DSDaniel Schneider
Daniel Schneider
EBEthan D. L. Brown D.L. Brown
Ethan D. L. Brown D.L. Brown
MWMax Ward
Max Ward
Daniel M. SciubbaDaniel M. Sciubba
Daniel M. Sciubba
SLSheng-fu Larry Lo
Sheng-fu Larry Lo
Healthcare systems increasingly deploy artificial intelligence to allocate resources, including procedure authorizations that impact patient access to care. While concerns about algorithmic bias typically focus on representation of protected attributes, how AI systems approach resource-constrained decisions remains understudied. We evaluated three large language models (LLMs), ChatGPT, Claude, and DeepSeek, on their handling of simulated surgical authorization request for an identical procedure. Each model assessed 6,500 surgeon profiles while implementing a mandated 30% denial rate, mirroring real-world authorization constraints. Multivariate regression analysis quantified how each model weighted 13 standardized attributes including professional qualifications and demographic characteristics. ChatGPT assigned significantly lower authorization scores to female surgeons (-9.55 points; 95% CI: -9.98, -9.11) while Claude (+ 2.01 points; 95% CI: + 1.85, + 2.17) and DeepSeek (+ 4.03 points; 95% CI: + 3.91, + 4.15) assigned higher scores to female surgeons. Geographic biases existed, with ChatGPT heavily favoring North American surgeons (+ 18.83 points; 95% CI: + 18.00, + 19.65) and DeepSeek penalizing them (-3.95 points; 95% CI: -4.18, -3.72). In ChatGPT, demographic factors frequently outweighed clinical qualifications; geographic location impacted authorization scores more than board certification. Though all models showed high internal consistency (R2 values 0.822–0.929), variability in prioritization of attributes resulted in divergent approval thresholds despite identical denial rates (ChatGPT: 64.6 ± 21.1, Claude: 68.5 ± 9.1, DeepSeek: 89.4 ± 9.2). We describe a phenomenon in AI healthcare decision-making which we term "constrained-resource divergence." When forced to discriminate between identical cases under resource constraints, AI systems may apply arbitrary weights that can impact patient care without clinical justification. In practice, this means patients with identical presentations may receive different authorization decisions based on which AI model their insurer deployed. Our findings raise profound questions about AI reliability for consequential healthcare decisions.

4 December 2025 in AI and Ethics
Verified
Journal Article
Open access
PDF file
DSDaniel Schneider
Daniel Schneider
EBEthan Devin Lockwood Brown
Ethan Devin Lockwood Brown
MWMax Ward
Max Ward
BOBarnabas Obeng-Gyasi
Barnabas Obeng-Gyasi
Daniel M. SciubbaDaniel M. Sciubba
Daniel M. Sciubba
SLSheng-fu Larry Lo
Sheng-fu Larry Lo
As artificial intelligence (AI) increasingly informs healthcare, understanding how large language models (LLMs) evaluate medical professionals is critical. This study quantified biases when LLMs assess neurosurgeon competency using demographic and practice characteristics. We prompted three prominent LLMs (ChatGPT-4o, Claude 3.7 Sonnet, and DeepSeek-V3) to score 6,500 synthetic neurosurgeon profiles. Profiles were created using demographically diverse names derived from public databases and randomly assigned professional attributes (experience, publications, institution, region, specialty) with statistical validation ensuring even distribution across groups. Multivariate regression analysis quantified how each factor influenced competency scores (0–100). Despite identical profiles, LLMs produced inconsistent mean (SD) scores: ChatGPT 91.85 (6.60), DeepSeek 71.74 (10.30), and Claude 62.29 (13.59). All models showed regional biases; North American neurosurgeons received scores 3.09 (ChatGPT) and 2.48 (DeepSeek) points higher than identical African counterparts ( P < .001). ChatGPT penalized East Asian (− 0.83), South Asian (− 0.91), and Middle Eastern (− 0.80) neurosurgeons ( P < .001). Practice setting bias was stronger, with ChatGPT and DeepSeek penalizing independent practitioners by 4.15 and 3.00 points, respectively, compared to hospital-employed peers ( P < .001). Models also displayed inconsistent bias correction, with ChatGPT elevating scores for female (+ 1.61) and Black-American (+ 1.69) neurosurgeons while disadvantaging other groups ( P < .001). This study provides evidence that LLMs incorporate distinct biases when evaluating neurosurgeons. As AI integration accelerates, uncritical adoption risks a self-reinforcing system where algorithmically preferred practitioners receive disproportionate advantages, independent of actual skills. These systems may also undermine global capacity-building by devaluing non-Western practitioners. Understanding and mitigating these biases is fundamental to responsibly navigating the intersection of medicine and AI.

30 January 2025 in Journal of Clinical Medicine
Verified
Journal Article
Open access
PDF file
MWMax Ward
Max Ward
DSDaniel Schneider
Daniel Schneider
EBEthan D. L. Brown D.L. Brown
Ethan D. L. Brown D.L. Brown
AMApratim Maity
Apratim Maity
Barnabas Obeng-GyasiBarnabas Obeng-Gyasi
Barnabas Obeng-Gyasi
SLSheng-fu Larry Lo
Sheng-fu Larry Lo
Background/Objectives: The increasing complexity of spinal oncology procedures, particularly in en-bloc tumor resections, creates challenges in tissue perfusion assessment due to extended operative times and extensive surgical dissection. Real-time visualization of tissue perfusion can be achieved with ICG using commercially available handheld imaging systems, offering potential advantages in spinal oncology cases. This study assessed the utility of ICG in analyzing soft-tissue viability during complex spine procedures extending beyond 7.5 h, with a particular focus on oncologic resections. Methods: Three cases that required over 7.5 h of operative time were chosen for ICG utilization. These cases included an en-bloc malignant peripheral nerve sheath tumor resection, an en-bloc resection of a malignant epithelioid neoplasm, and a long-segment fusion revision for pseudoarthrosis. At the conclusion of the critical portion of the procedure, a handheld intraoperative fluorescence camera was utilized to visualize the tissue penetration of intravenous ICG. Results: Prior to injecting ICG, devascularized tissue was not clearly visible. Injecting ICG allowed clear separation of vascularized (fluorescing) and devascularized (non-fluorescing) tissues. One region of non-florescent tissue was later confirmed to be devascularized with MRI and experienced postoperative infection. Conclusions: As the complexity of spinal oncology procedures increases, ICG fluorescence imaging offers a novel method for real-time assessment of tissue perfusion. This technique may be particularly valuable in extensive tumor resections, post-radiation cases, and revision surgeries where tissue viability is at risk. Further investigation in the spinal oncology population could help establish whether early identification of poorly perfused tissues impacts wound healing outcomes.