Basak Ozaslan

Background: Spinal anesthesia for cesarean delivery commonly causes maternal hypotension, which may compromise uteroplacental perfusion and maternal comfort. Guidelines recommend maintaining maternal blood pressure near baseline with prophylactic vasopressor strategies, yet titration remains reactive. We evaluated an autoregressive with exogenous input (ARX) decision-support algorithm that provides real-time forecasts of maternal mean arterial pressure (MAP) to support vasopressor management during cesarean delivery under spinal anesthesia. Methods: In this single-center, open-label, prospective pilot study, 20 pregnant patients at term undergoing elective cesarean delivery under spinal anesthesia received standard care supplemented by ARX-generated MAP predictions at 1-, 2- and 3-minute horizons. Clinicians titrated phenylephrine per institutional protocol while reviewing ARX predictions, retaining full autonomy for dosing decisions. Predictive performance was quantified using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and fraction of improvement in total error (FIT). ARX-guided patients were matched 1:2 to nonconcurrent controls (n = 40) on attending anesthesiologist and intrathecal bupivacaine dose, with nearest-neighbor matching on age and body mass index. Exploratory outcomes included hypotension (MAP Question: In pregnant patients at term undergoing elective cesarean delivery under spinal anesthesia, can a real-time ARX algorithm accurately forecast MAP and support vasopressor management? Findings: One-minute-ahead forecasts were accurate (RMSE 3.71 mmHg), and ARX-guided care was associated with a shorter duration of hypotension and a higher phenylephrine dose versus matched nonconcurrent controls Meaning: Real-time MAP forecasting is feasible and warrants randomized evaluation to confirm clinical benefit and characterize trade-offs.

Cesarean delivery under spinal anesthesia may be complicated by hypotension in up to 80% of the patients. The response to standard-of-care prophylactic phenylephrine infusion varies, and there is little guidance on achieving optimal blood pressure control. In this work, we developed a data-driven pharmacodynamic relationship between intravenous phenylephrine, intrathecal bupivacaine, and maternal mean arterial pressure (MAP) in patients presenting for cesarean delivery. In this single-center cohort study, secondary use data were available for normotensive patients presenting for cesarean delivery. Intraoperative MAP, intrathecal bupivacaine, and intravenous phenylephrine doses were recorded prospectively every minute. The recorded data were used to identify and confirm a time series (Autoregressive with Exogenous Input (ARX)) model for predicting the MAP using MATLAB 2021a System Identification Toolbox and the Prediction Error Method. An independent model validation was conducted using a second dataset collected after the model fitting stage. Model identification was performed on 172 patients, using 70% for model fitting and 30% for testing. The final ARX model, which takes the past three data points to make predictions, performed 48.9% better than a mean constant model for one-minute ahead MAP predictions with a root mean square error (RMSE) of 3.6 ± 1.3 mmHg. Similar performance was observed on independent validation using a second dataset (N = 84), yielding an RMSE of 4.2 ± 1.6 mmHg for one-minute ahead MAP predictions. Our ARX model showed good performance at up to a three-minute prediction horizon and could be used for future decision support applications to guide phenylephrine dose titration.

Objective: A common challenge in model identification with clinical data is incomplete and sometimes imprecise information. In this work, we provide a method to reconstruct the corrupted input data in a clinical dataset and, jointly identify the person-specific parameters of a metabolic model describing meal-insulin-glucose-dynamics for people with type 1 diabetes (T1D). Method: The proposed method is an algorithm that iterates between nonlinear least-squares and mixed-integer quadratic programming to optimize model parameters in conjunction with sparse corrections to the input data. In order to handle long stretches of data, the optimization problem is designed to be i) computationally tractable, and ii) robust against the potential presence of significant inaccuracies corrupting a data portion. Moreover, since the pattern of the inaccuracies is specific to each person, we propose a personalized hyperparameter tuning approach. The method is applied on clinical data from 13 people with T1D. Identified model performance is compared to the performance of model identified with standard least squares (LS) method. Results: Compared to LS, identifying corrections in conjunction with model parameters on training data lead to an improvement in the model prediction capabilities on unseen data with an average 2.2% improvement in MARD for two-hour prediction horizon (p-value = 0.0006). Conclusions: The proposed method is effective in model identification for clinical data with unknown inaccuracies in the inputs. Significance: Personalized models with high accuracy can inform treatment decisions and lead to better glucose control outcomes in people with T1D.

Objective: Evaluating the feasibility of closed-loop insulin delivery with a zone model predictive control (zone-MPC) algorithm designed for pregnancy complicated by type 1 diabetes (T1D). Research Design and Methods: Pregnant women with T1D from 14 to 32 weeks gestation already using continuous glucose monitor (CGM) augmented pump therapy were enrolled in a 2-day multicenter supervised outpatient study evaluating pregnancy-specific zone-MPC based closed-loop control (CLC) with the interoperable artificial pancreas system (iAPS) running on an unlocked smartphone. Meals and activities were unrestricted. The primary outcome was the CGM percentage of time between 63 and 140 mg/dL compared with participants' 1-week run-in period. Early (2-h) postprandial glucose control was also evaluated. Results: Eleven participants completed the study (age: 30.6 ± 4.1 years; gestational age: 20.7 ± 3.5 weeks; weight: 76.5 ± 15.3 kg; hemoglobin A1c: 5.6% ± 0.5% at enrollment). No serious adverse events occurred. Compared with the 1-week run-in, there was an increased percentage of time in 63–140 mg/dL during supervised CLC (CLC: 81.5%, run-in: 64%, P = 0.007) with less time >140 mg/dL (CLC: 16.5%, run-in: 30.8%, P = 0.029) and time P = 0.039). There was also less time 180 mg/dL (CLC: 4.9%, run-in: 13.1%, P = 0.032). Overnight glucose control was comparable, except for less time >250 mg/dL (CLC: 0%, run-in:3.9%, P = 0.030) and lower glucose standard deviation (CLC: 23.8 mg/dL, run-in:42.8 mg/dL, P = 0.007) during CLC. Conclusion: In this pilot study, use of the pregnancy-specific zone-MPC was feasible in pregnant women with T1D. Although the duration of our study was short and the number of participants was small, our findings add to the limited data available on the use of CLC systems during pregnancy (NCT04492566).

Background: Suboptimal glycemic control is associated with maternal and neonatal morbidity and mortality in pregnancy complicated by type 1 diabetes (T1D). Prospective analysis of continuous glucose monitoring (CGM) metrics, insulin pump settings, and insulin delivery can better characterize the changes in glycemic levels and insulin use throughout pregnancy with T1D. Materials and Methods: Prescribed parameters, insulin delivery, carbohydrate intake, and CGM data for 25 pregnant women with T1D from three U.S. sites were collected. Participants enrolled before 17 weeks gestation and used personal insulin pumps and study CGM. Mean daily total, basal, and bolus insulin doses (units/kg), CGM time in range (TIR: 63–140 mg/dL), and pump-entered carbohydrates were analyzed for every 2-week gestational interval. Linear mixed-effects regression models were used to evaluate changes across gestational ages compared to 12–14 weeks. Results: Basal insulin was higher during weeks 6–12 and 24–40. Daily bolus and total insulin were higher during weeks 20–40. Pump parameters were adjusted to intensify insulin therapy from 22 weeks onward. Average TIR across pregnancy was 59% ± 14%. Between 18 and 30 weeks, TIR was significantly lower, and time above range was significantly higher compared to the reference biweek. Time below target was lower between 22 and 34 weeks. Seven participants achieved >70% recommended TIR for pregnancy. Participants with maternal complications or infant neonatal intensive care unit admissions had lower TIR. Conclusion: While insulin dosing changed significantly with advancing gestation, most participants did not achieve >70% TIR. Customized anticipatory pump setting adjustments and automated systems aimed toward the designated TIR are needed to improve outcomes for this population. NCT03761615

Physical activity can cause glucose fluctuations both during and after it is performed, leading to hurdles in optimal insulin dosing in people with type 1 diabetes (T1D). We conducted a pilot clinical trial assessing the safety and feasibility of a physical activity-informed mealtime insulin bolus advisor that adjusts the meal bolus according to previous physical activity, based on step count data collected through an off-the-shelf physical activity tracker. Fifteen adults with T1D, each using a continuous glucose monitor (CGM) and an insulin pump with carbohydrate counting, completed two randomized crossover daily visits. Participants performed a 30 to 45-minute brisk walk before lunch and lunchtime insulin boluses were calculated based on either their standard therapy (ST) or the physical activity-informed bolus method. Post-lunch glycemic excursions were assessed using CGM readings. There was no significant difference between visits in the time spent in hypoglycemia in the post-lunch period (median [IQR] standard: 0 [0]% vs physical activity-informed: 0 [0]%, P = NS). Standard therapy bolus yielded a higher time spent in 70 to 180 mg/dL target range (mean ± standard: 77% ± 27% vs physical activity-informed: 59% ± 31%, P = .03) yet, it was associated with a steeper negative slope in the early postprandial phase (P = .032). Use of step count to adjust mealtime insulin following a walking bout has proved to be safe and feasible in a cohort of 15 T1D subjects. Physical activity-informed insulin dosing of meals eaten soon after a walking bout has a potential of mitigating physical activity related glucose reduction in the early postprandial phase.

Objective: In contrast with exercise, or structured physical activity (PA), glycemic disturbances due to daily unstructured PA in patients with type 1 diabetes (T1D) is largely underresearched, with limited information on treatment recommendations. We present results from retrospective analysis of data collected under patients' free-living conditions that illuminate the association between PA, as measured by an off-the-shelf activity tracker, and postprandial blood glucose control. Research Design and Methods: Data from 37 patients with T1D during two clinical studies with identical data collection protocols were analyzed retrospectively: 4 weeks of continuous glucose monitoring, carbohydrate intake, insulin injections, and PA (assessed through wearable activity tracker) were collected in free-living conditions. Five-hour glucose area under curves (GAUCs) following the last-bolused meal of every day were computed to assess postprandial glucose excursions, and their relation with corresponding antecedent PA was analyzed using linear mixed-effects regression models, accounting for meal, insulin, and current glycemic state. Results: Datasets yielded 845 days of data from 37 subjects (22.8 ± 11.6 days/subject); postmeal GAUC was negatively associated with total daily PA measured by step count ( P = 0.025), and total time spent performing higher than light-intensity PA ( P = 0.042). Patients with higher median total daily PA exhibited lower average postprandial GAUC ( P < 0.01). Additional analyses indicated that daily PA likely presents an immediate and delayed impact on glucose control. Conclusion: Daily PA assessed by commonly available sensors is significantly associated with glycemic exposure after an evening meal, indicating that quantitative assessment of PA may be useful in mealtime treatment decisions.

Objective: Suboptimal insulin dosing in type 1 diabetes (T1D) is frequently associated with time-varying insulin requirements driven by various psycho-behavioral and physiological factors influencing insulin sensitivity (IS). Among these, physical activity has been widely recognized as a trigger of altered IS both during and following the exercise effort, but limited indication is available for the management of structured and (even more) unstructured activity in T1D. In this work, we present two methods to inform insulin dosing with biosignals from wearable sensors to improve glycemic control in individuals with T1D. Research Design and Methods: Continuous glucose monitors (CGM) and activity trackers are leveraged by the methods. The first method uses CGM records to estimate IS in real time and adjust the insulin dose according to a person’s insulin needs; the second method uses step count data to inform the bolus calculation with the residual glucose-lowering effects of recently performed (structured or unstructured) physical activity. The methods were tested in silico within the University of Virginia/Padova T1D Simulator. A standard bolus calculator and the proposed “smart” systems were deployed in the control of one meal in presence of increased/decreased IS (Study 1) and following a 1-hour exercise bout (Study 2). Postprandial glycemic control was assessed in terms of time spent in different glycemic ranges and low/high blood glucose indices (LBGI/HBGI), and compared between the dosing strategies. Results: In Study 1, the CGM-informed system allowed to reduce exposure to hypoglycemia in presence of increased IS (percent time < 70 mg/dL: 6.1% versus 9.9%; LBGI: 1.9 versus 3.2) and exposure to hyperglycemia in presence of decreased IS (percent time > 180 mg/dL: 14.6% versus 18.3%; HBGI: 3.0 versus 3.9), tending toward optimal control. In Study 2, the step count-informed system allowed to reduce hypoglycemia (percent time < 70 mg/dL: 3.9% versus 13.4%; LBGI: 1.7 versus 3.2) at the cost of a minor increase in exposure to hyperglycemia (percent time > 180 mg/dL: 11.9% versus 7.5%; HBGI: 2.4 versus 1.5). Conclusions: We presented and validated in silico two methods for the smart dosing of prandial insulin in T1D. If seen within an ensemble, the two algorithms provide alternatives to individuals with T1D for improving insulin dosing accommodating a large variety of treatment options. Future work will be devoted to test the safety and efficacy of the methods in free-living conditions.