pradeep singh

Delays in identification of critical events in intensive care contribute to irreversible organ dysfunction and mortality(1). Shock Index(SI), calculated as the ratio of heart-rate and systolic blood pressure is a validated indicator for bedside examination, risk stratification, and prognostication(2,3). SI was originally designed for the evaluation of hemodynamic stability, with values >0.7 indicating shock. Shock is one of the highest contributors to higher mortality rates and SI has been used for the assessment of Shock in ICUs(4). The mortality rate in patients who develop shock in ICU is as high as 34% in developing countries(5). While in developing countries in Europe and North America the mortality rates are at 38% due to septic shock(6) Thus, prediction and early identification of Shock Index abnormality can help in initiating early life-saving therapies to prevent shock related mortality.(7). Early identification is critical for appropriate management(1), improved patient outcomes, and reduction of mortality(8–10).

Aditya Nagori, Pradeep Singh, Sameena Firdos, Vanshika Vats, Arushi Gupta, Harsh Bandhey, Anushtha Kalia, Arjun Sharma, Prakriti Ailavadi, Raghav Awasthi, Wrik Bhadra, Ayushmaan Kaul, Rakesh Lodha, Tavpritesh Sethi

Hypothermia is a life-threatening condition where the temperature of the body drops below 35°C and is a key source of concern in Intensive Care Units (ICUs). Early identification can help to nudge clinical management to initiate early interventions. Despite its importance, very few studies have focused on the early prediction of hypothermia. In this study, we aim to monitor and predict Hypothermia (30 min-4 h) ahead of its onset using machine learning (ML) models developed on physiological vitals and to prospectively validate the best performing model in the pediatric ICU. We developed and evaluated ML algorithms for the early prediction of hypothermia in a pediatric ICU. Sepsis advanced forecasting engine ICU Database (SafeICU) data resource is an in-house ICU source of data built in the Pediatric ICU at the All-India Institute of Medical Science (AIIMS), New Delhi. Each time-stamp at 1-min resolution was labeled for the presence of hypothermia to construct a retrospective cohort of pediatric patients in the SafeICU data resource. The training set consisted of windows of the length of 4.2 h with a lead time of 30 min-4 h from the onset of hypothermia. A set of 3,835 hand-engineered time-series features were calculated to capture physiological features from the time series. Features selection using the Boruta algorithm was performed to select the most important predictors of hypothermia. A battery of models such as gradient boosting machine, random forest, AdaBoost, and support vector machine (SVM) was evaluated utilizing five-fold test sets. The best-performing model was prospectively validated. A total of 148 patients with 193 ICU stays were eligible for the model development cohort. Of 3,939 features, 726 were statistically significant in the Boruta analysis for the prediction of Hypothermia. The gradient boosting model performed best with an Area Under the Receiver Operating Characteristic curve (AUROC) of 85% (SD = 1.6) and a precision of 59.2% (SD = 8.8) for a 30-min lead time before the onset of Hypothermia onset. As expected, the model showed a decline in model performance at higher lead times, such as AUROC of 77.2% (SD = 2.3) and precision of 41.34% (SD = 4.8) for 4 h ahead of Hypothermia onset. Our GBM(gradient boosting machine) model produced equal and superior results for the prospective validation, where an AUROC of 79.8% and a precision of 53% for a 30-min lead time before the onset of Hypothermia whereas an AUROC of 69.6% and a precision of 38.52% for a (30 min-4 h) lead time prospective validation of Hypothermia. Therefore, this work establishes a pipeline termed ThermoGnose for predicting hypothermia, a major complication in pediatric ICUs.

Pradeep Singh, Aditya Nagori, Tavpritesh Sethi

Shock is one of the major killers in intensive care units, and early interventions can potentially reverse it. In this study, we advance a noncontact thermal imaging modality for continuous monitoring of hemodynamic shock working on 1,03,936 frames from 406 videos recorded longitudinally upon 22 pediatric patients. Deep learning was used to preprocess and extract the Center-to-Peripheral Difference (CPD) in temperature values from the videos. This time-series data along with the heart rate was finally analyzed using Long-Short Term Memory models to predict the shock status up to the next 6 h. Our models achieved the best area under the receiver operating characteristic curve of 0.81 ± 0.06 and area under the precision-recall curve of 0.78 ± 0.05 at 5 h, providing sufficient time to stabilize the patient. Our approach, thus, provides a reliable shock prediction using an automated decision pipeline that can provide better care and save lives.

Vanshika Vats, Aditya Nagori, Pradeep Singh, Raman Dutt, Harsh Bandhey, Mahika Wason, Rakesh Lodha, Tavpritesh Sethi

Early prediction of hemodynamic shock in the ICU can save lives. Several studies have leveraged a combination of vitals, lab investigations, and clinical data to construct early warning systems for shock. However, these have a limited potential of generalization to diverse settings due to reliance on non-real-time data. Monitoring data from vitals can provide an early real-time prediction of Hemodynamic shock which can precede the clinical diagnosis to guide early therapy decisions. Generalization across age and geographical context is an unaddressed challenge. In this retrospective observational study, we built real-time shock prediction models generalized across age groups (adult and pediatric), ICU-types, and geographies. We trained, validated, and tested a shock prediction model on the publicly available eICU dataset on 208 ICUs across the United States. Data from 156 hospitals passed the eligibility criteria for cohort building. These were split hospital-wise in a five-fold training-validation-test set. External validation of the model was done on a pediatric ICU in New Delhi and MIMIC-III database with more than 0.23 million and one million patient-hours vitals data, respectively. Our models identified 92% of all the shock events more than 8 hours in advance with AUROC of 86 %(SD= 1.4) and AUPRC of 93% (SD =1.2) on the eICU testing set. An AUROC of 87 % (SD =1.8), AUPRC 92 % (SD=1.6) were obtained in external validation on the MIMIC-III cohort. The New Delhi Pediatric SafeICU data achieved an AUROC of 87 % (SD =4) AUPRC 91% (SD=3), despite being completely different geography and age group. In this first, we demonstrate a generalizable model for predicting shock, and algorithms are publicly available as a pre-configured Docker environment at https://github.com/tavlab-iiitd/ShoQPred.

Aditya Nagori, Pradeep Singh, Sameena Firdos, Vanshika Vats, Arushi Gupta, Harsh Bandhey, Anushtha Kalia, Arjun Sharma, Prakriti Ailavadi, Raghav Awasthi, Wrik Bhadra, Ayushmaan Kaul, Rakesh Lodha, Tavpritesh Sethi

Shock is a major killer in the ICU and Deep learning based early predictions can potentially save lives. Generalization across age and geographical context is an unaddressed challenge. In this retrospective observational study, we built real-time shock prediction models generalized across age groups and continents. More than 1.5 million patient-hours of novel data from a pediatric ICU in New Delhi and 5 million patient-hours from the adult ICU MIMIC database were used to build models. We achieved model generalization through a novel fractal deep-learning approach and predicted shock up to 12 hours in advance. Our deep learning models showed a receiver operating curve (AUROC) drop from 78% (95%CI, 73-83) on MIMIC data to 66% (95%CI, 54-78) on New Delhi data, outperforming standard machine learning by nearly a 10% gap. Therefore, better representations and deep learning can partly address the generalizability-gap of ICU prediction models trained across geographies. Our data and algorithms are publicly available as a pre-configured docker environment at https://github.com/SAFE-ICU/ShoQPred.

Aditya Nagori, Anushtha Kalia, Arjun Sharma, Pradeep Singh, Harsh Bandhey, Prakriti Ailavadi, Raghav Awasthi, Wrik Bhadra, Ayushmaan Kaul, Rakesh Lodha, Tavpritesh Sethi

The relationship between meteorological factors such as temperature and humidity with COVID-19 incidence is still unclear after 6 months of the beginning of the pandemic. Some literature confirms the association of temperature with disease transmission while some oppose the same. This work intends to determine whether there is a causal association between temperature, humidity and Covid-19 cases. Three different causal models were used to capture stochastic, chaotic and symbolic natured time-series data and to provide a robust & unbiased analysis by constructing networks of causal relationships between the variables. Granger-Causality method, Transfer Entropy method & Convergent Cross-Mapping (CCM) was done on data from regions with different temperatures and cases greater than 50,000 as of 13th May 2020. From the Granger-Causality test we found that in only Canada, the United Kingdom, temperature and daily new infections are causally linked. The same results were obtained from Convergent Cross Mapping for India. Again using Granger-Causality test, we found that in Russia only, relative humidity is causally linked to daily new cases. Thus, a Generalized Additive Model with a smoothing spline function was fitted for these countries to understand the directionality. Using the combined results of the said models, we were able to conclude that there is no evidence of a causal association between temperature, humidity and Covid-19 cases.

Raghav Awasthi, Aditya Nagori, Pradeep Singh, Ridam Pal, Vineet Joshi, Tavpritesh Sethi

The flood of conflicting COVID-19 research has revealed that COVID-19 continues to be an enigma. Although more than 14,000 research articles on COVID-19 have been published with the disease taking a pandemic proportion, clinicians and researchers are struggling to distill knowledge for furthering clinical management and research. In this study, we address this gap for a targeted user group, i.e. clinicians, researchers, and policymakers by applying natural language processing to develop a CovidNLP dashboard in order to speed up knowledge discovery. The WHO has created a repository of about more than 5000 peer-reviewed and curated research articles on varied aspects including epidemiology, clinical features, diagnosis, treatment, social factors, and economics. We summarised all the articles in the WHO Database through an extractive summarizer followed by an exploration of the feature space using word embeddings which were then used to visualize the summarized associations of COVID-19 as found in the text. Clinicians, researchers, and policymakers will not only discover the direct effects of COVID-19 but also the systematic implications such as the anticipated rise in TB and cancer mortality due to the non-availability of drugs during the export lockdown as highlighted by our models. These demonstrate the utility of mining massive literature with natural language processing for rapid distillation and knowledge updates. This can help the users understand, synthesize, and take pre-emptive action with the available peer-reviewed evidence on COVID-19. Our models will be continuously updated with new literature and we have made our resource CovidNLP publicly available in a user-friendly fashion at http://covidnlp.tavlab.iiitd.edu.in/.

Raghav Awasthi, Ridam Pal, Harshita Chopra, Harsh Bandhey, Pradeep Singh, Aditya Nagori, Suryatej Reddy, Amogh Gulati, Ponnurangam Kumaraguru, Tavpritesh Sethi

The flood of conflicting COVID-19 research has revealed that COVID-19 continues to be an enigma. Although more than 14,000 research articles on COVID-19 have been published with the disease taking a pandemic proportion, clinicians and researchers are struggling to distill knowledge for furthering clinical management and research. In this study, we address this gap for a targeted user group, i.e. clinicians, researchers, and policymakers by applying natural language processing to develop a CovidNLP dashboard in order to speed up knowledge discovery. The WHO has created a repository of about more than 5000 peer-reviewed and curated research articles on varied aspects including epidemiology, clinical features, diagnosis, treatment, social factors, and economics. We summarised all the articles in the WHO Database through an extractive summarizer followed by an exploration of the feature space using word embeddings which were then used to visualize the summarized associations of COVID-19 as found in the text. Clinicians, researchers, and policymakers will not only discover the direct effects of COVID-19 but also the systematic implications such as the anticipated rise in TB and cancer mortality due to the non-availability of drugs during the export lockdown as highlighted by our models. These demonstrate the utility of mining massive literature with natural language processing for rapid distillation and knowledge updates. This can help the users understand, synthesize, and take pre-emptive action with the available peer-reviewed evidence on COVID-19. Our models will be continuously updated with new literature and we have made our resource CovidNLP publicly available in a user-friendly fashion at http://covidnlp.tavlab.iiitd.edu.in/.

Raghav Awasthi, Ridam Pal, Harshita Chopra, Harsh Bandhey, Pradeep Singh, Aditya Nagori, Suryatej Reddy, Amogh Gulati, Ponnurangam Kumaraguru, Tavpritesh Sethi

Shock is one of the major killers in ICUs and early interventions can potentially reverse it. In this study, we advance a non-contact thermal imaging modality to continuous monitoring of hemodynamic shock working on 406 patient videos of 256 seconds length for 22 patients longitudinally. Deep learning was performed upon these videos to extract Center-to-Peripheral Difference (CPD) in temperature values. CPD along with heart rate, was finally analysed to predict the shock status up to next 12 hours using Long-Short Term Memory models. Our models achieved best area under the receiver-operating-characteristics curve of 0.81 ± 0.06 and area under precision-recall curve of 0.78 ± 0.05 at 5 hours, providing sufficient time to stabilize the patient. Our approach, thus, provides a reliable prediction using an automated decision pipeline, that can save lives and provide better care.

Vanshika Vats, Pradeep Singh, Aditya Nagori, Raman Dutt, Harsh Bandhey, Mahika Wason, Rakesh Lodha, Tavpritesh Sethi