Microneedle-Array-Electrode-Based ECG with PPG Sensor for Cuffless Blood Pressure Estimation

In this talk I will overview the Medical Information Mart for Intensive Care (MIMIC) database, a continuing series of publicly available critical care databases. As of 2021, the latest iteration of MIMIC is MIMIC-IV, which contains 12 years of admissions to critical care units at the Beth Israel Deaconess Medical Center in Boston, MA. MIMIC has also expanded into other domains, including imaging with the MIMIC Chest X-Ray database (MIMIC-CXR), and the MIMIC-IV Emergency Department database (MIMIC-IV-ED). I will discuss our motivations in distributing MIMIC to the world, the challenges in publicly releasing medical data, and the impact the databases have had on the community. A crucial aspect of the safe and privacy preserving use of retrospective clinical data is deidentification, where patient identifiers are removed or replaced with unidentifiable surrogates. I will cover approaches we have created for deidentifying patient clinical data, images, and free-text clinical notes. I’ll conclude with my thoughts regarding the future of healthcare data modeling, including the need for novel data structures which ensure data quality and utility for broad research use.

An awakening in medicine: the partnership of humanity and intelligent machines

The key to cuffless blood pressure (BP) estimation lies in identifying the efficient noninvasive features or signals that can indicate BP changes. Most these features are identified based on their correlation with BP. However, the correlation does not imply causation, as there might be confounders impacting the relationship between the features and BP. In this study, we introduce causal inference to identify the confounders that can possibly affect the accuracy of the commonly studied pulse transit time (PTT) based cuffless BP estimation. We further propose a hierarchical regression model to address the impact of the confounder thereby improving the estimation performance. We first build the causal graph of BP and common features extracted from electrocardiogram (ECG) and photoplethysmogram (PPG) signals with the fast causal inference algorithm and causal generative neural network. We found heart rate is the main confounder that impacting the relationship between PTT and BP. To consider the effect of the confounder, we then developed a hierarchical regression model for cuffless BP estimation. We validated the proposed method on 17 subjects, and the results show that the mean absolute difference (MAD) of the estimated systolic BP and diastolic BP were 6.29 mmHg and 5.03 mmHg respectively, which significantly outperformed the linear regression model without the hierarchical strategy.

Accurate cuffless blood pressure (BP) estimation remains challenging, particularly under dynamic conditions with significant intra-individual BP variations. This study introduces UTransBPNet, a novel, calibration-free model for cuffless BP estimation. It integrates a squeeze-and-excitation-enhanced Unet architecture for short-range feature extraction with a transformer and cross attention module to capture long-range dependencies from high-resolution, multi-channel physiological signals, further refined through an optimized fine-tuning scheme. Comprehensive validations were conducted across multiple dynamic datasets—Dataset_Drink, Dataset_Exercise, and Dataset_MIMIC—in both scenario-specific and cross-scenario settings. Results demonstrate that UTransBPNet outperformed existing models in tracking BP variations under dynamic conditions, achieving individual Pearson’s correlation coefficients of 0.61 ± 0.17 and 0.62 ± 0.13 for systolic BP (SBP) and diastolic BP (DBP) in Dataset_Drink, 0.82 ± 0.11 and 0.72 ± 0.18 in Dataset_Exercise, and low mean absolute differences (MADs) of 4.38 and 2.25 mmHg in Dataset_MIMIC. The analysis also highlights the impact of dataset characteristics on model performance, such as distribution shift, distribution imbalance and individual BP variability, highlighting the need for well-curated data to ensure generalizability. This study advances the development of robust, cuffless BP estimation models for real-world applications.

Objective: Our study aims to evaluate the association between heart rate variability (HRV) and short- and long-term prognosis in patients admitted to intensive care unit (ICU). Methods and Results: Adult patients continuously monitored for over 24 h in ICUs from the the American Medical Information Mart for Intensive Care (MIMIC)-IV Waveform Database were recruited in our study. Twenty HRV-related variables (8 time domain, 6 frequency domain, and 6 nonlinear variables) were calculated based on RR intervals. The association between HRV and all-cause mortality was assessed. Ninety-three patients met the inclusion criteria and were classified into atrial fibrillation (AF) and sinus rhythm (SR) groups, which were further divided into 30-day survivor group and nonsurvivor\groups based on their survival status. The 30-day all-cause mortality rates in AF and SR groups were 36.3% and 14.6%, respectively. All the time domain, frequency domain, and nonlinear HRV parameters did not differ significantly between survivors and nonsurvivors with or without AF (all P > 0.05). Presence of renal failure, malignancy, and elevated blood urea nitrogen level were associated with increased 30-day all-cause mortality in SR patients, while presence of sepsis, infection, higher platelet count, and magnesium level were associated with increased 30-day all-cause mortality in AF patients. Conclusions: Heart rate variability variables were not associated with increased 30-day all-cause mortality in ICU patients with or without AF.

ObjectiveThis study explored the association between serum bicarbonate levels and mortality risk among patients with acute respiratory distress syndrome (ARDS) admitted to the intensive care unit (ICU).MethodsThis was a retrospective cohort study utilizing data extracted from the Medical Information Mart for Intensive Care (MIMIC)-IV database. Cox proportional hazards models and restricted cubic splines (RCS) were deployed for elucidating the association between the baseline bicarbonate levels and the risk of 28-day mortality while utilizing the Kaplan-Meier method to estimate survival curves, with hazard ratio (HR) and 95% confidence interval (CI). Subgroup analyses were conducted based on age, gender, Charlson Comorbidity Index (CCI) score, ARDS severity and bicarbonate administration.ResultsTotally, 6,377 patients (15.38% deaths) were included. Baseline bicarbonate was significantly associated with 28-day mortality (HR: 0.98, 95% CI: 0.97–1.00, P = 0.011) in patients with ARDS. This association was particularly evident in female patients (HR: 1.16, 95% CI: 1.14–1.87, P = 0.003), those with a CCI of 2 or higher (HR: 1.27, 95% CI: 1.05–1.53, P = 0.013), among those with a PaO2/FiO2 ratio ranging from 200 to 300 mmHg (HR: 1.39, 95% CI: 1.08–1.78, P = 0.011), and those without bicarbonate administration (HR = 1.26, 95%CI: 1.07–1.48, P = 0.004), where bicarbonate levels falling below 23 mEq/L were linked to a heightened risk of not surviving the first 28 days in ARDS patients. RCS analysis revealed that the bicarbonate levels were non-linear associated with the 28-day mortality in ARDS patients (P for non-linear <0.001).ConclusionLower serum bicarbonate levels are significantly associated with an increased 28-day mortality risk in ARDS patients, with particular emphasis on female patients, those with higher CCI scores, and those with milder ARDS. Baseline bicarbonate levels of ARDS patients in ICU have certain clinical reference value for the development of clinical management and the assessment of prognostic risk during the ICU admission.

Severe acute pancreatitis (SAP) has a mortality of 30% with no current targeted therapy. The potential protective effect of insulin on AP has been reported and needs to be confirmed. Thus, we aim to examine the effect of insulin treatment on the outcome of AP patients. A retrospective study was performed using data from the Medical Information Mart for Intensive Care (MIMIC) database. Kruskal-Wallis test, t-tests, and Pearson's chi-squared test were used to compare differences between groups. Propensity score matching and further nearest neighbor matching were used to construct a matched cohort. Cox proportional hazards regression analyses, logistic regression analyses, and the doubly robust estimation method were used to assess the relationship between insulin use and mortality. Nine hundred patients were enrolled in the final analysis. Insulin was associated with better outcomes in AP patients admitted to ICU, and could act as an independent predictor for 30-day mortality (HR = 0.36, 95% CI = 0.24-0.55). Subgroup analysis showed that AP patients with heart failure or without kidney disease or respiratory failure may not benefit from insulin treatment. Insulin treatment is independently associated with lower 30-day mortality in AP patients, except for those with heart failure or without kidney disease or respiratory failure.

Although blood glucose changes have been suggested to be a potential better target for clinical control than baseline blood glucose levels, the association of blood glucose changes with the prognosis in acute myocardial infarction (AMI) patients with diabetes mellitus (DM) is unclear. Herein, this study aimed to investigate association of short-term longitudinal trajectory of blood glucose with 30-day mortality in this population. Data of AMI patients with DM were extracted from the Medical Information Mart for Intensive Care (MIMIC) database in 2003-2019 in this retrospective cohort study. The latent growth mixture modeling (LGMM) model was utilized to classify the 24-hour longitudinal trajectory of blood glucose of the patients. Kaplan-Meier (KM) curve was drawn to show 30-day mortality risk in patients with different trajectory classes. Univariate and multivariate Cox regression analyses were employed to explore the association of longitudinal trajectory of blood glucose within 24 hours after the ICU admission with 30-day mortality. Also, subgroups analysis of age, gender, and AMI types was performed. The evaluation indexes were hazard ratios (HRs) and 95% confidence intervals (CIs). Among 1,523 eligible patients, 227 (14.9%) died within 30 days. We identified 4 longitudinal trajectories of blood glucose, including class 1 (a low initial average blood glucose level with steady trend within 24 hours), class 2 (a high initial average blood glucose with gently decreased trend), class 3 (the highest initial average blood glucose with rapidly decreased trend) and class 4 (a high initial average blood glucose level with the trend that increased at first and then decreased). After adjusting for covariates, an average blood glucose level of ≥200 mg/dL was linked to higher risk of 30-day mortality, comparing to that of <140 mg/dL (HR = 1.80, 95%CI: 1.23-2.63). Comparing to patients whose longitudinal trajectory of blood glucose conformed to class 1, those with class 2 (HR = 2.52, 95%CI: 1.79-3.53) or class 4 (HR = 3.53, 95%CI: 2.07-6.03) seemed to have higher risk of 30-day mortality. Additionally, these associations were also significant in aged ≥60 years old, female, male, NSTEMI, and STEMI subgroups (all P<0.05). A low level of average blood glucose at the ICU admission or reducing blood glucose to a normal level quickly with adequate measures in 24 hours after ICU admission may be beneficial for AMI patients with DM to reduce the risk of 30-day mortality. These findings may provide some information for further exploration on appropriate range of blood glucose changes in clinical practice.

BackgroundAcute kidney injury (AKI) in patients with acute myocardial infarction (AMI) often indicates a poor prognosis.ObjectiveThis study aimed to investigate the association between the TyG index and the risk of AKI in patients with AMI.MethodsData were taken from the Medical Information Mart for Intensive Care (MIMIC) database. A 1:3 propensity score (PS) was set to match patients in the AKI and non-AKI groups. Multivariate logistic regression analysis, restricted cubic spline (RCS) regression and subgroup analysis were performed to assess the association between TyG index and AKI.ResultsTotally, 1831 AMI patients were included, of which 302 (15.6%) had AKI. The TyG level was higher in AKI patients than in non-AKI patients (9.30 ± 0.71 mg/mL vs. 9.03 ± 0.73 mg/mL, P < 0.001). Compared to the lowest quartile of TyG levels, quartiles 3 or 4 had a higher risk of AKI, respectively (Odds Ratiomodel 4 = 2.139, 95% Confidence Interval: 1.382–3.310, for quartile 4 vs. quartile 1, Ptrend < 0.001). The risk of AKI increased by 34.4% when the TyG level increased by 1 S.D. (OR: 1.344, 95% CI: 1.150–1.570, P < 0.001). The TyG level was non-linearly associated with the risk of AKI in the population within a specified range. After 1:3 propensity score matching, the results were similar and the TyG level remained a risk factor for AKI in patients with AMI.ConclusionHigh levels of TyG increase the risk of AKI in AMI patients. The TyG level is a predictor of AKI risk in AMI patients, and can be used for clinical management.

BackgroundThe Endothelial Activation and Stress Index (EASIX) has recently gained attention as an emerging indicator of endothelial injury, which significantly contributes to the progression of sepsis. Nevertheless, the association between EASIX levels and mortality outcomes in sepsis has yet to be fully clarified. This study explored the clinical significance of EASIX in septic patients admitted to the intensive care unit (ICU).MethodsData of sepsis patients were acquired from the Medical Information Mart for Intensive Care (MIMIC) database and analyzed. Participants were categorized into three groups according to their log2–EASIX values. The interplay between the mortality and log2–EASIX values in individuals with sepsis was examined via Cox regression analysis. To explore possible nonlinear relationships between log2–EASIX values and sepsis-related mortality, we applied a restricted cubic spline (RCS) model. Additionally, Kaplan–Meier (KM) and subgroup analyses were conducted to evaluate the robustness of the findings.ResultsA total of 8829 individuals diagnosed with sepsis were included in this study. Elevated log2–EASIX levels were consistently associated with a higher likelihood of mortality throughout all assessed time frames. Patients with the highest log2–EASIX scores exhibited a markedly elevated risk of death, with hazard ratios (HRs) of 1.56 at 30 days, 1.44 at 60 days, 1.38 at 90 days, 1.31 at 180 days, and 1.29 at 365 days. While the association was strongest in the early phase, elevated mortality risk persisted throughout the follow-up period. Moreover, an alternative statistical approach, specifically the use of RCS, exhibited a steady linear link between log2–EASIX values and mortality risk across all considered time points.ConclusionsThis study demonstrated that EASIX is independently and positively associated with mortality in patients with sepsis. Elevated EASIX levels were correlated with increased disease severity and poorer clinical outcomes, suggesting its potential utility as a risk stratification indicator in critical care.

To introduce Medical Information Mart for Intensive Care (MIMIC) database and elaborate the approach of critically emergent research with big data based on the feature of MIMIC and updated studies both domestic and overseas, we put forward the feasibility and necessity of introducing medical big data to research in emergency. Then we discuss the role of MIMIC database in emergency clinical study, as well as the principles and key notes of experimental design and implementation under the medical big data circumstance. The implementation of MIMIC database in emergency medical research provides a brand new field for the early diagnosis, risk warning and prognosis of critical illness, however there are also limitations. To meet the era of big data, emergency medical database which is in accordance with our national condition is needed, which will provide new energy to the development of emergency medicine.

To evaluate the efficacy of therapeutic hypothermia in enhancing cognitive and motor outcomes six months post-TBI. A retrospective cohort analysis was conducted on data from the Medical Information Mart for Intensive Care (MIMIC) database. A total of 259 TBI patients were divided into a Routine group (n = 135) and a Hypothermia Therapy group (n = 124). The hypothermia group received core temperature reduction to 33-35°C for 3-5 days. Cognitive and motor functions were assessed using the Functional Independence Measure (FIM), Fugl-Meyer Motor Assessment (FMA), and Glasgow Coma Scale (GCS) at baseline and six months post-treatment. Patients' satisfaction was also evaluated. After six months of treatment, the hypothermia group showed significantly higher FIM motor scores, and FIM cognitive scores than the routine group (P < 0.01). A higher proportion of patients in the hypothermia group achieved mild impairment on the FMA assessment than in the routine group (P = 0.003). The hypothermia group demonstrated better consciousness levels (P = 0.006) and reported significantly higher patient satisfaction rate (P = 0.004). Hypothermia therapy significantly improved cognitive and motor recovery in TBI patients, enhanced consciousness levels, and increased patient satisfaction rate.

BackgroundThe relationship between serum sodium levels and stroke prognosis has been investigated, but the effect of serum sodium fluctuations on severe stroke patients remains uncertain. We aimed to evaluate the association between serum sodium trajectories within 72 h of admission and all-cause mortality in intensive care unit (ICU) patients with ischemic stroke.MethodsThis retrospective study utilized data from the Medical Information Mart for Intensive Care (MIMIC) database, including patients diagnosed with ischemic stroke from 2008 to 2022. Serum sodium trajectories within 72 h of admission were classified using a latent class trajectory model. Kaplan–Meier survival curves and Cox proportional hazards models were employed to assess the relationship between trajectories and all-cause mortality. Subgroup analyses were conducted to assess the consistency of this association across various populations.ResultsAmong 881 enrolled patients, two distinct serum sodium trajectory groups were identified: class 1 (n = 778, 88.3%) exhibited stable sodium levels within the normal range, while class 2 (n = 103, 11.7%) showed rapid fluctuations. Kaplan–Meier analyses revealed that patients in class 2 had higher all-cause mortality at 30-day, 90-day, 180-day, and 1-year follow-ups (all P < 0.001). Multivariate COX regression analyses consistently showed elevated mortality risk in class 2 versus class 1 across all timepoints: ICU (HR: 2.17; 95% CI 1.27–3.68; P = 0.004), in-hospital (HR: 2.51; 95% CI 1.64–3.85; P < 0.001), 30-day (HR: 2.41; 95% CI 1.67–3.49; P < 0.001), 90-day (HR: 2.17; 95% CI 1.56–3.03; P < 0.001), 180-day (HR: 2.13; 95% CI 1.55–2.92; P < 0.001) and 1 year (HR: 1.99; 95% CI 1.46–2.71; P < 0.001). Subgroup analyses confirmed the robustness of these findings.ConclusionsIn critically ill patients with ischemic stroke, rapidly fluctuating serum sodium levels are independently associated with increased short-term and long-term all-cause mortality. Further investigations are needed to elucidate the role of sodium dynamics in stroke outcomes.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40001-025-03320-w.

Objective To investigate the association between preoperative nutritional scores and moderate-to-severe acute kidney injury (AKI) after coronary artery bypass graft (CABG) surgery and the predictive significance of nutritional indices for moderate to severe AKI. Methods This study retrospectively included older patients underwent CABG surgery from the Medical Information Mart for Intensive Care (MIMIC) database. Nutritional scores were calculated by the Geriatric Nutritional Risk Index (GNRI) and the Prognostic Nutritional Index (PNI), respectively. Moderate-to-severe injury was determined by KDIGO criteria. Logistic regression, subgroup analysis, and restricted cubic splines were utilized to investigate the association. The predictive value was also assessed by the area under the curve (AUC), net reclassification index (NRI), and integrated discrimination improvement (IDI). Results A total of 1,007 patients were retrospectively included, of which 100 (9.9%) and 380 (37.7%) had malnutrition calculated by GNRI and PNI scores. The incidence of moderate-to-severe AKI was 524 (52.0%). After adjustment for selected risk factors, worse nutritional scores were associated with a higher incidence of moderate-to-severe AKI (P GNRI<0.001; P PNI=0.001). Integrating these indices into different base models improves their performance, as manifested by significant improvements in AUCs and NRIs (p < 0.05). Conclusion Worse preoperative nutritional status was associated with an elevated risk of postoperative moderate-to-severe AKI. Integrating these indices into base models improve their predictive performance. These results highlight the importance of assessing nutritional status among older patients had CABG surgery.

PurposeThis study was aimed to identify the risk factors that influence the mortality risk in patients with acute aortic dissection (AAD) within one year after discharge, and aimed to construct a predictive model for assessing mortality risk.MethodsThe study involved 320 adult patients obtained from the Medical Information Mart for Intensive Care (MIMIC) database. Logistic regression analysis was conducted to identify potential risk factors associated with mortality in AAD patients within one year after discharge and to develop a predictive model. The performance of the predictive model was assessed using the receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA). To further validate the findings, patient data from the First Affiliated Hospital of Guangxi Medical University (157 patients) were analyzed.ResultsUnivariate and multivariate logistic regression analyses revealed that gender, length of hospital stay, highest blood urea nitrogen (BUN_max), use of adrenaline, and use of amiodarone were significant risk factors for mortality within one year after discharge (p < 0.05). The constructed model exhibited a consistency index (C-index) and an area under the ROC curve of 0.738. The calibration curve and DCA demonstrated that these indicators had a good degree of agreement and utility. The external validation results of the model also indicated good predictability (AUC = 0.700, p < 0.05).ConclusionThe personalized scoring prediction model constructed by gender, length of hospital stays, BUN_max levels, as well as the use of adrenaline and amiodarone, can effectively identify AAD patients with high mortality risk within one year after discharge.