Interleukin-6 improves infection identification when added to physician judgment during evaluation of potentially septic patients.

BACKGROUND: Identifying infection is critical in early sepsis screening. This study assessed whether biomarkers of endothelial activation and/or inflammation could improve identification of infection among Emergency Department (ED) patients with organ dysfunction. METHODS: We performed a prospective, observational study at two urban, academic EDs, between June 2016 and December 2017. We included admitted adults with 1) two systemic inflammatory response syndrome criteria and organ dysfunction, 2) systolic blood pressure < 90 mmHg, or 3) lactate ≥4.0 mmol/L. We excluded patients with trauma, transferred for intracranial hemorrhage, or without available blood samples. Treating ED physicians reported presence of infection (yes/no) at inpatient admission. Assays for angiopoietin-1, angiopoietin-2, soluble tumor necrosis factor receptor-1, interleukin-6, and interleukin-8 were performed using ED blood samples. The primary outcome was infection, adjudicated by paired physician review. Using logistic regression, we compared the performance of physician judgment, biomarkers, and physician judgment-biomarkers combination to predict infection. Area under the curve (AUC) and AUC 95% confidence intervals were estimated by bootstrap procedure. RESULTS: Of 421 patients enrolled, 306 patients met final study criteria. Of these, 154(50.3%) patients had infectious etiologies. Physicians correctly discriminated infectious from non-infectious etiologies in 239 (78.1%). Physician judgment performed moderately when discriminating infection (AUC 0.78, 95% CI: 0.74-0.82) and outperformed the best biomarker model, interleukin-6 alone, (AUC 0.71, 0.66-0.76). Physician judgment improved when including interleukin-6 (AUC 0.84, 0.79-0.87), with modest AUC improvement: 0.06 (0.03-0.08). CONCLUSIONS: In ED patients with organ dysfunction, plasma interleukin-6 may improve infection discrimination when added to physician judgment.

Factors Impacting Physician Prognostic Accuracy in Heart Failure Patients With Reduced Left Ventricular Ejection Fraction.

BACKGROUND: A recent study showed that the accuracy of heart failure (HF) cardiologists and family doctors to predict mortality in outpatients with HF proved suboptimal, performing less well than models. OBJECTIVES: The authors sought to evaluate patient and physician factors associated with physician accuracy. METHODS: The authors included outpatients with HF from 11 HF clinics. Family doctors and HF cardiologists estimated patient 1-year mortality. They calculated predicted mortality using the Seattle HF Model and followed patients for 1 year to record mortality (or urgent heart transplant or ventricular assist device implant as mortality-equivalent events). Using multivariable logistic regression, the authors evaluated associations among physician experience and confidence in estimates, duration of patient-physician relationship, patient-physician sex concordance, patient race, and predicted risk, with concordant results between physician and model predictions. RESULTS: Among 1,643 patients, 1-year event rate was 10% (95% CI: 8%-12%). One-half of the estimates showed discrepant results between model and physician predictions, mainly owing to physician risk overestimation. Discrepancies were more frequent with increasing patient risk from 38% in low-risk to ∼75% in high-risk patients. When making predictions on male patients, female HF cardiologists were 26% more likely to have discrepant predictions (OR: 0.74; 95% CI: 0.58-0.94). HF cardiologist estimates in Black patients were 33% more likely to be discrepant (OR: 0.67; 95% CI: 0.45-0.99). Low confidence in predictions was associated with discrepancy. Analyses restricted to high-confidence estimates showed inferior calibration to the model, with risk overestimation across risk groups. CONCLUSIONS: Discrepant physician and model predictions were more frequent in cases with perceived increased risk. Model predictions outperform physicians even when they are confident in their predictions. (Predicted Prognosis in Heart Failure [INTUITION]; NCT04009798).

Confidence Calibration: An Introduction With Application to Quality Improvement.

A probabilistic forecast is one that assigns a probability (or likelihood) to the occurrence of an event. Radiologists commonly make probabilistic judgments in their reports, even if these predictions are not explicitly expressed as numbers. There are calls for radiologists to commit to their probabilistic predictions in a standardized fashion; however, without a mechanism for feedback, there is no opportunity for improvement. Analysis techniques familiar to radiologists (eg, calculation of sensitivity and specificity and construction of receiver operating characteristics curves) have a blind spot with regard to calibration of these probabilities to reality and are the main obstacle to improvement along this axis. We review statistical and graphical methods for calibration analysis in wider use outside the medical literature and present a framework for implementation of these techniques for quality improvement and radiologist self-assessment.

Against the iDoctor: why artificial intelligence should not replace physician judgment.

Experts in medical informatics have argued for the incorporation of ever more machine-learning algorithms into medical care. As artificial intelligence (AI) research advances, such technologies raise the possibility of an "iDoctor," a machine theoretically capable of replacing the judgment of primary care physicians. In this article, I draw on Martin Heidegger's critique of technology to show how an algorithmic approach to medicine distorts the physician-patient relationship. Among other problems, AI cannot adapt guidelines according to the individual patient's needs. In response to the objection that AI could develop this capacity, I use Hubert Dreyfus's analysis of AI to argue that attention to the needs of each patient requires the physician to attune his or her perception to the patient's history and physical exam, an ability that seems uniquely human. Human physician judgment will remain better suited to the practice of primary care despite anticipated advances in AI technology.