Publications

We develop an unsupervised probabilistic model for heterogeneous Electronic Health Record (EHR) data. Utilizing a mixture model formulation, our approach directly models sequences of arbitrary length, such as medications and laboratory results. This allows for subgrouping and incorporation of the dynamics underlying heterogeneous data types. The model consists of a layered set of latent variables that encode underlying structure in the data. These variables represent subject subgroups at the top layer, and unobserved states for sequences in the second layer. We train this model on episodic data from subjects receiving medical care in the Kaiser Permanente Northern California integrated healthcare delivery system. The resulting properties of the trained model generate novel insight from these complex and multifaceted data. In addition, we show how the model can be used to analyze sequences that contribute to assessment of mortality likelihood.

Abstract Despite demonstrated facility for arithmetic and other quantitative tasks, the performance of ChatGPT and other large language models for clinical risk calculation have yet to be assessed. Using synthetic patient data, this preliminary study aimed to assess the calibration, reproducibility, and potential for sociodemographic bias of ChatGPT-derived Pooled Cohort Equation (PCE) scores of atherosclerotic cardiovascular disease risk as compared to true scores. We found that ChatGPT-derived PCE scores, despite being moderately associated with the true PCE scores, displayed poor calibration with respect to true PCE scores, and exhibited instability between repeated rounds of prompting, suggesting lack of reproducibility. Moreover, ChatGPT-derived PCE scores also appeared inappropriately sensitive to contextual indicators of the sociodemographic status of the synthetic patients in this study. Further work is needed to confirm these results, and to assess performance on a wider variety of prompts as well as in other settings beyond cardiovascular disease prevention where accurate risk calculation is also vital to appropriate clinical decision-making. Abstract Figure Figure. Underestimation of true PCE risk estimates ( x -axis) by ChatGPT ( y -axis) on synthetic patient data.

Sepsis is the main cause of multiple organ failure and remains a concern because of the associated high morbidity and mortality. In recent years, important advances have been made in the understanding of the pathophysiology of sepsis. Sepsis and septic shock are the end result of complex interactions between infecting organisms and various elements of the host response. A key feature of the common sequence of organ failure is dysfunction of the cardiovascular system, including microcirculatory elements. Outcome improvement in sepsis is based on recognizing the process early and instituting effective therapies. The time window for intervention is relatively short, and treatment must promptly control the source of infection, restore haemodynamic homoeostasis, and support failing organ systems.

Because the formation of platelet thrombi has been incriminated in the development of the adult respiratory distress syndrome (ARDS), we tested the hypothesis that early administration of antiplatelet agents might protect the lungs in patients at risk to develop ARDS after circulatory shock. This double-blinded study included 40 patients treated with either 3 mg/kg . 24 h of dipyridamole or a corresponding dose of placebo after an episode of hemorrhagic, traumatic, or septic shock. Each patient also received 100 mg of aspirin daily. Arterial blood gases and chest x-rays were not significantly different between dipyridamole and placebo groups. Moreover, two patients receiving dipyridamole but none receiving placebo developed ARDS. This pilot study does not support a beneficial effect of dipyridamole in the prevention of ARDS after circulatory shock.

Shock is a state of ‘acute circulatory failure’, the key feature of which is an inability for tissues and cells to get enough oxygen to meet their needs, ultimately resulting in cell death. Shock can be classified as hypovolemic, cardiogenic, obstructive or distributive although many patients will have several types of shock simultaneously. Although it is important to identify and treat the underlying cause of shock (e.g., antibiotics and source removal for septic shock; thrombolysis or embolectomy for massive pulmonary embolism causing obstructive shock, hemodynamic support must be started immediately in all cases to provide a minimum perfusion pressure and prevent development or worsening of organ dysfunction. In this context, both “flow” and “pressure” are important components. Indeed, the arterial pressure is determined by blood flow and vascular tone, i.e., blood pressure = cardiac output x systemic vascular resistance. The essential aspects of shock resuscitation can be remembered using the simple VIP mnemonic: ventilate (ensure adequate oxygenation), infuse (provide adequate fluid resuscitation), and pump (administer vasoactive agents). Fluid administration should be guided by repeated fluid challenges so that patients receive enough fluid but not too much, as excess fluid can have multiple harmful effects. If hypotension is severe, vasopressors should be started early, at the same time as fluids, to increase systemic vascular resistance and thus arterial pressure. Prolonged periods of hypotension are associated with worse outcomes. Importantly, although an initial mean arterial pressure (MAP) target of 65 mmHg may be a useful aim, this will not be optimal for all and target values should be adapted according to the individual patient, taking into account various factors including age and history of chronic hypertension. Indeed, if the MAP target is too low, resultant hypoperfusion may lead to cellular death and organ dysfunction, but a target that is too high may be associated with edema and excessive vasoconstriction as a result of higher amounts of fluid and vasoactive agents, which may also impair organ function. Patients with circulatory shock must therefore be carefully monitored, including regular assessment of cardiac output, and treatment and targets adapted accordingly. Monitoring organ perfusion at the bedside is difficult without specific tools to assess the microcirculation. As such, we must generally rely on three “windows” that can indicate inadequate perfusion, i.e., impaired cutaneous perfusion, impaired renal function, and impaired mental status. Plasma lactate levels can also be useful, with changes over time providing some indication of adequacy of tissue oxygenation. Although these changes are too slow to help acutely guide therapy, the trend can provide valuable information about patient status over time. If flow remains inadequate and there is no, or only a poor, response to fluids, an inotropic agent may be considered. Dobutamine is the inotrope of choice. In this context, measurement of mixed (SvO 2 ) or central venous (ScvO 2 ) oxygen saturation can help as it provides an indication of the balance between oxygen delivery and consumption, with low values ( < 70%) suggesting that increasing oxygen delivery could be beneficial.

Increased awareness of the signs and symptoms of sepsis and an emphasis on the importance of early treatment have helped to improve survival rates from this serious and frequent condition in recent years. With no specific, effective anti-sepsis therapies available, management focuses on early source control with adequate and appropriate antibiotics and removal of any source of infection, rapid resuscitation, hemodynamic stabilization and organ support. Use of dedicated teams to care for patients with sepsis can help optimize early management.