Hidden Flaws Behind Expert-Level Accuracy of Multimodal GPT-4 Vision in Medicine.

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges - an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Impact of preoperative endoscopic procedures on adverse event rates after surgical resection for main-duct and mixed-type intraductal papillary mucinous neoplasms (IPMNs).

BACKGROUND: Main-duct (MD-) and mixed-type (MT-) IPMNs harbor an increased risk of pancreatic cancer and warrant surgical resection. Preoperative endoscopic retrograde cholangiopancreatography (ERCP) and endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) are important in the diagnosis of IPMNs. The aim of this study was to investigate whether endoscopic procedures manipulating the MD impact postoperative adverse events in patients with MD- and MT-IPMNs. METHODS: We performed a retrospective study of 369 patients who underwent resections for MD- or MT-IPMN at two tertiary centers (2000-2019). Multivariable logistic regression analyses were performed for postoperative adverse events to compare the risks between intervention (ERCP, EUS-FNA with branch duct (BD) aspirated, EUS-FNA with MD aspirated from the duct directly or cyst/mass arising from MD) versus no-intervention group. RESULTS: 33.1 % of patients had a preoperative ERCP and 69.4 % had EUS-FNA. Postoperative adverse events included: 30-day readmission (12.7 %), delayed gastric emptying (13.8 %), pancreatic fistula (10.3 %), abdominal abscess (5.7 %), cardiopulmonary adverse events (11.4 %), and mortality (1.4 %). The model was adjusted for potential confounders. There were no significant differences between the ERCP and no-ERCP groups for specific adverse events. Compared to no-EUS-FNA groups, groups of EUS-FNA with BD aspiration and EUS-FNA with MD aspiration from the main pancreatic duct directly or cyst/mass arising from MD did not show a significant increase in specific adverse events. CONCLUSIONS: Postoperative adverse events were not significantly increased among patients who had ERCP or EUS-FNA before surgical resection for MD- or MT-IPMNs. Endoscopic procedures directly sampling the MD can be safely pursued for diagnostic purposes in selected cases.

Elucidating the Drivers for the Rising Incidence of Early-Onset Colorectal Cancer: How Ecologic Studies Could Help and What Is Next.

The incidence of colorectal cancer diagnosed before age 50, often referred to as early-onset colorectal cancer, has been increasing, whereas the overall colorectal cancer incidence has declined. Elucidating the drivers for the rising burden of early-onset colorectal cancer is a priority in cancer epidemiology and prevention. In this issue of Cancer Epidemiology, Biomarkers & Prevention, Chen and colleagues demonstrated that ecologic studies are a helpful method to reveal emerging risk factors at the population level and concluded that alcohol use might be a potential contributor to the rising incidence of early-onset colorectal cancer. Moving forward, because of the observed birth cohort effect in early-onset colorectal cancer, where younger generations have a steeper increase, hypothesis-driven investigations on emerging risk factors in recent generations, especially during early life, are warranted. Ultimately, the identified risk factors could be integrated with well-established microsimulation models of colorectal cancer, powerful tools that can simultaneously capture population-level secular changes in risk factors, relative risk estimates for each risk factor, and the natural history of colorectal cancer. This would allow us to quantitatively estimate the explained and unexplained portion of the rising incidence of early-onset colorectal cancer by calendar period and birth cohorts, and to help identify priorities in etiologic research, prevention, and early detection. See related article by Chen et al., p. 217.

Total ice content and lipid saturation determine adipose tissue cryolipolysis by injection of ice-slurry.

OBJECTIVES: Cryolipolysis uses tissue cooling to solidify lipids, preferentially damaging lipid-rich cells. Topical cooling is popular for the reduction of local subcutaneous fat. Injection of biocompatible ice-slurry is a recently introduced alternative. We developed and verified a quantitative model that simulates the heat exchange and phase changes involved, offering insights into ice-slurry injection for treating subcutaneous fat. METHODS: Finite element method was used to model the spatial and temporal progression of heat transfer between adipose tissue and injected ice-slurry, estimating dose-response relationships between properties of the slurry and size of tissue affected by cryolipolysis. Phase changes of both slurry and adipose tissue lipids were considered. An in vivo swine model was used to validate the numerical solutions. Oils with different lipid compositions were exposed to ice-slurry in vitro to evaluate the effects of lipid freezing temperature. Microscopy and nuclear magnetic resonance (NMR) were performed to detect lipid phase changes. RESULTS: A ball of granular ice was deposited at the injection site in subcutaneous fat. Total injected ice content determines both the effective cooling region of tissue, and the duration of tissue cooling. Water's high latent heat of fusion enables tissue cooling long after slurry injection. Slurry temperature affects the rate of tissue cooling. In swine, when 30 ml slurry injection at -3.5°C was compared to 15 ml slurry injection at -4.8°C (both with the same total ice content), the latter led to almost twice faster tissue cooling. NMR showed a large decrease in diffusion upon lipid crystallization; saturated lipids with higher freezing temperatures were more susceptible to solidification after ice-slurry injection. CONCLUSIONS: Total injected ice content determines both the volume of tissue treated by cryolipolysis and the cooling duration after slurry injection, while slurry temperature affects the cooling rate. Lipid saturation, which varies with diet and anatomic location, also has an important influence.

Association of Proton Pump Inhibitor Use With All-Cause and Cause-Specific Mortality.

BACKGROUND & AIMS: The use of proton pump inhibitors (PPIs) has increased rapidly in the past 2 decades. Concerns about the regular use of PPIs contributing to mortality have been raised. METHODS: We conducted a prospective cohort study using data collected from the Nurses' Health Study (2004-2018) and the Health Professionals Follow-up Study (2004-2018). Cox proportional hazards models were used to estimate the hazard ratios (HRs) and 95% CIs for mortality according to PPI use. We used a modified lag-time approach to minimize reverse causation (ie, protopathic bias). RESULTS: Among 50,156 women and 21,731 men followed for 831,407 person-years and a median of 13.8 years, we documented 22,125 deaths, including 4592 deaths from cancer, 5404 from cardiovascular diseases, and 12,129 deaths from other causes. Compared with nonusers of PPIs, PPI users had significantly higher risks of all-cause mortality (HR, 1.19; 95% CI, 1.13-1.24) and mortality due to cancer (HR, 1.30; 95% CI, 1.17-1.44), cardiovascular diseases (HR, 1.13; 95% CI, 1.02-1.26), respiratory diseases (HR, 1.32; 95% CI, 1.12-1.56), and digestive diseases (HR, 1.50; 95% CI, 1.10-2.05). Upon applying lag times of up to 6 years, the associations were attenuated and no longer statistically significant (all-cause: HR, 1.04; 95% CI, 0.97-1.11; cancer: HR, 1.07; 95% CI, 0.89-1.28; cardiovascular diseases: HR, 0.94; 95% CI, 0.81-1.10; respiratory diseases: HR, 1.20; 95% CI, 0.95-1.50; digestive diseases: HR, 1.38; 95% CI, 0.88-2.18). Longer duration of PPI use did not confer higher risks for all-cause and cause-specific mortality. CONCLUSIONS: After accounting for protopathic bias, PPI use was not associated with higher risks of all-cause mortality and mortality due to major causes.

Feeding intolerance in critically ill patients with COVID-19.

BACKGROUND & AIMS: Early reports suggest significant difficulty with enteral feeding in critically ill COVID-19 patients. This study aimed to characterize the prevalence, clinical manifestations, and outcomes of feeding intolerance in critically ill patients with COVID-19. METHODS: We examined 323 adult patients with COVID-19 admitted to the intensive care units (ICUs) of Massachusetts General Hospital between March 11 and June 28, 2020 who received enteral nutrition. Systematic chart review determined prevalence, clinical characteristics, and hospital outcomes (ICU complications, length of stay, and mortality) of feeding intolerance. RESULTS: Feeding intolerance developed in 56% of the patients and most commonly manifested as large gastric residual volumes (83.9%), abdominal distension (67.2%), and vomiting (63.9%). Length of intubation (OR 1.05, 95% CI 1.03-1.08), ≥1 GI symptom on presentation (OR 0.76, 95% CI 0.59-0.97), and severe obesity (OR 0.29, 95% CI 0.13-0.66) were independently associated with development of feeding intolerance. Compared to feed-tolerant patients, patients with incident feeding intolerance were significantly more likely to suffer cardiac, renal, hepatic, and hematologic complications during their hospitalization. Feeding intolerance was similarly associated with poor outcomes including longer ICU stay (median [IQR] 21.5 [14-30] vs. 15 [9-22] days, P < 0.001), overall hospitalization time (median [IQR] 30.5 [19-42] vs. 24 [15-35], P < 0.001) and in-hospital mortality (33.9% vs. 16.1%, P < 0.001). Feeding intolerance was independently associated with an increased risk of death (HR 3.32; 95% CI 1.97-5.6). CONCLUSIONS: Feeding intolerance is a frequently encountered complication in critically ill COVID-19 patients in a large tertiary care experience and is associated with poor outcomes.

Subcutaneous Fat Reduction with Injected Ice Slurry.

BACKGROUND: Cryolipolysis is a noninvasive method for removal of subcutaneous fat for body contouring. Conventional cryolipolysis with topical cooling requires extracting heat from subcutaneous fat by conduction across the skin, thus limiting the amount and the location of the fat removed. The authors hypothesized that local injection of a physiological ice slurry directly into target adipose tissue would lead to more efficient and effective cryolipolysis. METHODS: Injectable slurries containing 20 percent and 40 percent ice content were made using common parenteral agents (normal saline and glycerol), then locally injected into the subcutaneous fat of swine. Ultrasound imaging, photography, histological, and gross tissue responses were monitored before and periodically up to 8 weeks after injection. RESULTS: Fat loss occurred gradually over several weeks following a single ice slurry injection. There was an obvious and significant 55 ± 6 percent reduction in adipose tissue thickness compared with control sites injected with the same volume of melted slurry (p < 0.001, t test). The amount of fat loss correlated with the total volume of ice injected. There was no scarring or damage to surrounding tissue. CONCLUSION: Physiological ice slurry injection is a promising new strategy for selective and nonsurgical fat removal.

Antiepileptic drug treatment after an unprovoked first seizure: A decision analysis.

OBJECTIVE: To compare the expected quality-adjusted life-years (QALYs) in adult patients undergoing immediate vs deferred antiepileptic drug (AED) treatment after a first unprovoked seizure. METHODS: We constructed a simulated clinical trial (Markov decision model) to compare immediate vs deferred AED treatment after a first unprovoked seizure in adults. Three base cases were considered, representing patients with varying degrees of seizure recurrence risk and effect of seizures on quality of life (QOL). Cohort simulation was performed to determine which treatment strategy would maximize the patient's expected QALYs. Sensitivity analyses were guided by clinical data to define decision thresholds across plausible measurement ranges, including seizure recurrence rate, effect of seizure recurrence on QOL, and efficacy of AEDs. RESULTS: For patients with a moderate risk of recurrent seizures (52.0% over 10 years after first seizure), immediate AED treatment maximized QALYs compared to deferred treatment. Sensitivity analyses showed that for the preferred choice to change to deferred AED treatment, key clinical measures needed to reach implausible values were 10-year seizure recurrence rate ≤38.0%, QOL reduction with recurrent seizures ≤0.06, and efficacy of AEDs on lowering seizure recurrence rate ≤16.3%. CONCLUSION: Our model determined that immediate AED treatment is preferable to deferred treatment in adult first-seizure patients over a wide and clinically relevant range of variables. Furthermore, our analysis suggests that the 10-year seizure recurrence rate that justifies AED treatment (38.0%) is substantially lower than the 60% threshold used in the current definition of epilepsy.

[Protective effect of Danxuetong injection against testicular injury after testis torsion/detorsion in rats].

OBJECTIVE: To investigate the protective effect of Danxuetong injection (DXT, a combination of Danshen and Xueshuantong injections) against testicular ischemia-reperfusion injury following testis torsion/detorsion in rats. METHODS: Thirty-two 4-week-old healthy male SD rats were randomly divided into four groups of equal number: sham operation, normal saline, single DXT injection, and successive DXT injection. The rat models of testicular ischemia-reperfusion injury were established by 2-hour 720-degree torsion/detorsion of the unilateral testis. At 6 weeks after modeling, the rats were killed and their testes were harvested for measure- ment of testicular coefficients, sperm counts, sperm motility, and the levels of total anti-oxidative capacity (T-AOC) , superoxide dismutase (SOD) , nitric oxide synthase (NOS) , and malondialdehyde ( MDA) in the testis tissue. RESULTS: Compared with the rats of the normal saline group, those of the single DXT injection and successive DXT injection groups showed significant increases in the testicular coefficient (0.11 ± 0.03 vs 0.35 ± 0.04 and 0.40 ± 0.06, P < 0.05), sperm count ([0.46 ± 0.10] vs [1.44 ± 0.50] and [3.00 ± 1.28] x10(9)/ml, P < 0.05), sperm motility ([13.63 ± 14.04] vs [39.63 ± 5.04] and [76.31 ± 3.67]%, P < 0.05), the activity of SOD (72.76 ± 5.58 vs 116.25 ± 8.83 and 133.20 ± 13.84, P < 0.05), and the level of T-AOC (5.58 ± 1.07 vs 13.34 ± 5.81 and 19.21 ± 5.69, P < 0.05), but a remarkable decrease in the content of MDA (42.38 ± 8.94 vs 20.94 ± 5.65 and 15.02 ± 1.03, P < 0. 05) in the injured testes. CONCLUSION: DXT can effectively rid the testis tissue of oxygen free radicals, improve sperm count and motility by antioxidation, and protect the testis tissue of prepubertal rats against testicular ischemia-reperfusion injury after testis torsion/detorsion. It also has a protective effect on the contralateral testis, and successive injection has a better effect than single injection of DXT.