Ajan Subramanian

Ajan Subramanian

Auto-generated profile, based on publicly available information

Ajan Subramanian

Department of Computer Science, University of California, Irvine, Irvine, CA, United States

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
28 May 2021 in Journal of Medical Internet Research
Verified
Journal Article
Open access
ENEmad Kasaeyan Naeini
Emad Kasaeyan Naeini
ASAjan Subramanian
Ajan Subramanian
MCMichael-David Calderon
Michael-David Calderon
KZKai Zheng
Kai Zheng
NDNikil Dutt
Nikil Dutt
ARAmir Rahmani
Amir Rahmani
Background: There is a strong demand for an accurate and objective means of assessing acute pain among hospitalized patients to help clinicians provide pain medications at a proper dosage and in a timely manner. Heart rate variability (HRV) comprises changes in the time intervals between consecutive heartbeats, which can be measured through acquisition and interpretation of electrocardiography (ECG) captured from bedside monitors or wearable devices. As increased sympathetic activity affects the HRV, an index of autonomic regulation of heart rate, ultra–short-term HRV analysis can provide a reliable source of information for acute pain monitoring. In this study, widely used HRV time and frequency domain measurements are used in acute pain assessments among postoperative patients. The existing approaches have only focused on stimulated pain in healthy subjects, whereas, to the best of our knowledge, there is no work in the literature building models using real pain data and on postoperative patients. Objective: The objective of our study was to develop and evaluate an automatic and adaptable pain assessment algorithm based on ECG features for assessing acute pain in postoperative patients likely experiencing mild to moderate pain. Methods: The study used a prospective observational design. The sample consisted of 25 patient participants aged 18 to 65 years. In part 1 of the study, a transcutaneous electrical nerve stimulation unit was employed to obtain baseline discomfort thresholds for the patients. In part 2, a multichannel biosignal acquisition device was used as patients were engaging in non-noxious activities. At all times, pain intensity was measured using patient self-reports based on the Numerical Rating Scale. A weak supervision framework was inherited for rapid training data creation. The collected labels were then transformed from 11 intensity levels to 5 intensity levels. Prediction models were developed using 5 different machine learning methods. Mean prediction accuracy was calculated using leave-one-out cross-validation. We compared the performance of these models with the results from a previously published research study. Results: Five different machine learning algorithms were applied to perform a binary classification of baseline (BL) versus 4 distinct pain levels (PL1 through PL4). The highest validation accuracy using 3 time domain HRV features from a BioVid research paper for baseline versus any other pain level was achieved by support vector machine (SVM) with 62.72% (BL vs PL4) to 84.14% (BL vs PL2). Similar results were achieved for the top 8 features based on the Gini index using the SVM method, with an accuracy ranging from 63.86% (BL vs PL4) to 84.79% (BL vs PL2). Conclusions: We propose a novel pain assessment method for postoperative patients using ECG signal. Weak supervision applied for labeling and feature extraction improves the robustness of the approach. Our results show the viability of using a machine learning algorithm to accurately and objectively assess acute pain among hospitalized patients. International Registered Report Identifier (IRRID): RR2-10.2196/17783

27 January 2025 in JMIR Formative Research
Verified
Journal Article
Open access
ASAjan Subramanian
Ajan Subramanian
RCRui Cao
Rui Cao
ENEmad Kasaeyan Naeini
Emad Kasaeyan Naeini
SASeyed Amir Hossein Aqajari
Seyed Amir Hossein Aqajari
THThomas D Hughes
Thomas D Hughes
ARAmir Rahmani
Amir Rahmani
Background: Acute pain management is critical in postoperative care, especially in vulnerable patient populations that may be unable to self-report pain levels effectively. Current methods of pain assessment often rely on subjective patient reports or behavioral pain observation tools, which can lead to inconsistencies in pain management. Multimodal pain assessment, integrating physiological and behavioral data, presents an opportunity to create more objective and accurate pain measurement systems. However, most previous work has focused on healthy subjects in controlled environments, with limited attention to real-world postoperative pain scenarios. This gap necessitates the development of robust, multimodal approaches capable of addressing the unique challenges associated with assessing pain in clinical settings, where factors like motion artifacts, imbalanced label distribution, and sparse data further complicate pain monitoring. Objective: This study aimed to develop and evaluate a multimodal machine learning–based framework for the objective assessment of pain in postoperative patients in real clinical settings using biosignals such as electrocardiogram, electromyogram, electrodermal activity, and respiration rate (RR) signals. Methods: The iHurt study was conducted on 25 postoperative patients at the University of California, Irvine Medical Center. The study captured multimodal biosignals during light physical activities, with concurrent self-reported pain levels using the Numerical Rating Scale. Data preprocessing involved noise filtering, feature extraction, and combining handcrafted and automatic features through convolutional and long-short-term memory autoencoders. Machine learning classifiers, including support vector machine, random forest, adaptive boosting, and k-nearest neighbors, were trained using weak supervision and minority oversampling to handle sparse and imbalanced pain labels. Pain levels were categorized into baseline and 3 levels of pain intensity (1-3). Results: The multimodal pain recognition models achieved an average balanced accuracy of over 80% across the different pain levels. RR models consistently outperformed other single modalities, particularly for lower pain intensities, while facial muscle activity (electromyogram) was most effective for distinguishing higher pain intensities. Although single-modality models, especially RR, generally provided higher performance compared to multimodal approaches, our multimodal framework still delivered results that surpassed most previous works in terms of overall accuracy. Conclusions: This study presents a novel, multimodal machine learning framework for objective pain recognition in postoperative patients. The results highlight the potential of integrating multiple biosignal modalities for more accurate pain assessment, with particular value in real-world clinical settings.

12 June 2023 in medRxiv
Verified
Preprint
Open access
PDF file
ASAjan Subramanian
Ajan Subramanian
RCRui Cao
Rui Cao
ENEmad Kasaeyan Naeni
Emad Kasaeyan Naeni
SASyed Amir Hossein Aqajari
Syed Amir Hossein Aqajari
THThomas D. Hughes
Thomas D. Hughes
ARAmir Rahmani
Amir Rahmani
Objective To develop and evaluate a multimodal machine learning-based objective pain assessment algorithm on data collected from post-operative patients. Methods The proposed method addresses the major challenges that come with using data from such patients like the imbalanced distribution of pain classes and the scarcity of ground-truth labels. Specifically, we extracted automatic features using a convolutional autoencoder (AE) along with data augmentation techniques like weak supervision and minority oversampling to improve our models’ predictive performance. This method was used in conjunction with four different machine learning classifiers: Adaptive Boosting (AdaBoost), Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) to perform binary classification on three increasing levels of pain when compared to no pain. Results Our models are able to recognize different pain levels with an average balanced accuracy of over 80%. Conclusion This is the first multimodal pain recognition work done on postoperative patients and our proposed method provides valuable insights for automatic acute pain recognition in such patients.

23 March 2023 in medRxiv
Verified
Preprint
Open access
PDF file
ASAjan Subramanian
Ajan Subramanian
YHYong Huang
Yong Huang
MPMelissa D. Pinto
Melissa D. Pinto
CDCharles A. Downs
Charles A. Downs
ARAmir Rahmani
Amir Rahmani
Traditional machine learning (ML) approaches learn to recognize patterns in the data but fail to go beyond observing associations. Such data-driven methods can lack generalizability when the data is outside the independent and identically distributed (i.i.d) setting. Using causal inference can aid data-driven techniques to go beyond learning spurious associations and frame the data-generating process in a causal lens. We can combine domain expertise and traditional ML techniques to answer causal questions on the data. Hypothetical questions on alternate realities can also be answered with such a framework. In this paper, we estimate the causal effect of Pre-Exposure Prophylaxis (PrEP) on mortality in COVID-19 patients from an observational dataset of over 120,000 patients. With the help of medical experts, we hypothesize a causal graph that identifies the causal and non-causal associations, including the list of potential confounding variables. We use estimation techniques such as linear regression, matching, and machine learning (meta-learners) to estimate the causal effect. On average, our estimates show that taking PrEP can result in a 2.1% decrease in the death rate or a total of around 2,540 patients’ lives saved in the studied population.

30 September 2026 in JMIR Formative Research
Verified
Journal Article
Open access
YHYong Huang
Yong Huang
NNNitish Nagesh
Nitish Nagesh
ASAjan Subramanian
Ajan Subramanian
LWLi Wang
Li Wang
ZLZhiyu Liu
Zhiyu Liu
RSRandall Stafford
Randall Stafford
Background: Suboptimal glycemic control remains a significant public health challenge among adults with type 2 diabetes and prediabetes, with 47.4% of US adults with diagnosed diabetes having hemoglobin A 1c (HbA 1c ) ≥7%. Remote glucose monitoring programs have shown promise for supporting self-management, but real-world evidence on the causal impact of varying patient engagement levels on glycemic outcomes remains limited. Objective: This study aimed to estimate the causal dose-response relationship between patient engagement, operationalized as weekly glucose monitoring frequency, and glycemic control measured by HbA 1c among adults enrolled in a comprehensive primary care–integrated remote monitoring program. Methods: We conducted a retrospective cohort study of 1436 adults with type 2 diabetes or prediabetes enrolled in the Unika Health program between 2019 and 2024. The program integrated Bluetooth-connected glucose meters, a mobile app, structured lifestyle coaching, and primary care coordination across 74 physician practices. Engagement was defined as mean weekly glucose monitoring frequency during the first 6 months. The causal effect of monitoring frequency on 6-month HbA 1c was estimated using marginal structural models (MSMs) with inverse probability weighting to address time-varying confounding. Covariates included age, sex, BMI, baseline HbA 1c , comorbidities, medication status, and physical activity level. Results: The cohort was predominantly older (95.4%, 1370/1436 aged ≥46 years), with 82.6% (1186/1436) having hypertension and 45.9% (659/1436) classified as obese. Overall, HbA 1c decreased by a mean of 0.54 (SD 1.47; 95% CI 0.47‐0.62) percentage points ( P <.001) over 6 months. Cluster analysis identified 3 engagement tiers: low (n=835; mean 2.55, SD 1.35 measurements/week), medium (n=493; mean 6.19, SD 1.46 measurements/week), and high (n=108; mean 12.59, SD 3.03 measurements/week). A monotonic dose-response was observed, with mean HbA 1c reductions of 0.38 (SD 1.40; 95% CI 0.29‐0.48), 0.71 (SD 1.50; 95% CI 0.58‐0.85), and 1.01 (SD 1.67; 95% CI 0.69‐1.32) percentage points for the low, medium, and high tiers, respectively (all P <.001 by 2-tailed paired t test). In weighted MSMs, each additional weekly measurement was associated with a 0.05 (95% CI 0.03‐0.07) percentage point greater HbA 1c reduction ( P <.001). Among patients with high baseline HbA 1c (≥9%), the high-engagement group achieved a mean reduction of 3.12 (SD 1.97; 95% CI 2.29‐3.94) percentage points. Findings were consistent in sensitivity analyses at 3 months (β=−0.04; P <.001) and 12 months (β=−0.03; P =.003) and across alternative weighting specifications. Conclusions: Higher engagement with a digitally enabled, primary care–integrated remote glucose monitoring program was causally associated with significantly greater HbA 1c reductions in adults with type 2 diabetes and prediabetes. These findings support scalable remote patient monitoring strategies that actively foster sustained patient engagement as an effective approach to improving glycemic control and reducing the burden of diabetes-related complications at a population level.