BMBella Mehta

Bella Mehta

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Bella Mehta has a total of 164 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Artificial Intelligence‐Driven Extraction of Patient‐Reported Pain Outcomes in Osteoarthritis Using Large Language Models

    by Jainesh Doshi, Stephen Batter, Yiyuan Wu, William Salerno, Alice Santilli, Sandhya Kannayiram, Susan Goodman, Bella Mehta

    Objective: Patient‐reported outcome measures (PROMs) are central to assessing symptom burden and treatment response in Osteoarthritis (OA) but are often embedded in unstructured clinical notes. We evaluated whether prompt‐engineered large language models (LLMs) could accurately extract or infer pain scores on a 0‐10 scale from routine OA clinical documentation. Methods: We retrospectively analyzed 163 paired internal medicine and orthopedic clinic notes from patients in a prospective OA registry. A trained abstractor extracted explicitly documented pain scores from notes. For notes lacking explicit pain scores, physicians estimated pain severity from narrative documentation. The Llama 3.2‐90B model, using Chain‐of‐Thought and Few‐Shot prompting, inferred pain scores. Registry‐collected pain scores served as a reference standard. Agreement was assessed using Kendall's Tau‐b. Results: Agreement between LLM‐extracted and abstracted scores was strong for IM notes (n =132, τ = 0.76, p < 0.0001) and excellent for orthopedic notes (n = 65, τ = 0.95, p < 0.0001). The LLM identified 15 EMR‐generated default zero scores in IM notes; after excluding them, agreement was perfect (τ = 1.0, p < 0.0001). For notes without explicit pain scores (31 IM and 98 orthopedic notes), LLM‐based pain inference showed weak correlation with registry scores (IM: τ = 0.17; orthopedic: τ = 0.11), but performed comparably to physician estimation. LLM–physician agreement ranged from 54.2%–88.0% within ±1 point and 66.7%–100% within ±2 points for IM notes, and from 54.8%–70.5% within ±1 point and 84.5%–97.7% within ±2 points for orthopedic notes. Conclusion: LLMs accurately extract explicitly documented PROMs from clinical notes and reliably identify documentation artifacts. Inference from narrative documentation was less precise when notes lacked sufficient contextual information and detailed pain descriptions. image

  • Using Artificial Intelligence in Qualitative Lupus Pregnancy Research: The Potential of Large Language Models and Prompting Strategies

    by Julia Romina M. Boers, Grace Terry, Larissa Rodrigues, Fernanda Surita, Jainesh Doshi, Bella Mehta

    Objective: Our objective was to evaluate the ability of different large language models (LLMs) to translate and analyze qualitative interviews describing pregnancy experiences among women with systemic lupus erythematosus (SLE). An additional objective was to compare themes identified by LLMs with those identified by human researchers. Methods: We analyzed 26 semistructured interviews originally conducted in Portuguese by Rodrigues et al among women with SLE exploring their perinatal care experiences. Three prompting strategies (zero‐shot, few‐shot, and chain‐of‐thought) were used on the transcripts across four state‐of‐the‐art LLMs (GPT‐4o, OpenAI; Claude Sonnet 4.5, Anthropic; Gemini 2.5 Flash, Google; and DeepSeek R, DeepSeek) to generate key themes. Model outputs were evaluated using an LLM‐as‐a‐judge framework assessing thematic coverage, thematic unity, and semantic similarity. Results: Across models and prompting strategies, LLMs generated similar themes that closely aligned with those identified by human researchers. LLM‐as‐a‐judge evaluations demonstrated high thematic unity and semantic similarity. Claude Sonnet 4.5 showed the most consistent performance. Few‐shot prompting consistently outperformed zero‐shot and chain‐of‐thought approaches. Conclusion: LLMs show promise as supportive tools for analyzing qualitative research by preserving thematic coherence and capturing the complexity of pregnancy experiences in women with SLE. Although zero‐shot prompting is widely used, it produced lower thematic results in our study, whereas few‐shot prompting achieved the most comprehensive and coherent theme extraction. LLM‐assisted qualitative analysis may complement human interpretation and has the potential to support the inclusion of non–English language data and diverse patient perspectives in lupus research. image

  • Clinical and Histopathologic Characteristics and Outcomes of Patients With Psoriatic Arthritis Undergoing Arthroplasty

    by Josh Mukherjee, Amit Lakhanpal, Melanie H. Smith, Laura T. Donlin, Jose A. Rodriguez, Peter K. Sculco, Mark P. Figgie, Jason L. Blevins, Daniel C. Ramirez, Edward F. DiCarlo, et al.

    Objective: Patients with psoriatic arthritis (PsA) experience higher postoperative complication rates after total hip or total knee arthroplasty (THA and TKA, respectively) than patients with osteoarthritis, but contributing factors to this risk are poorly defined. We examined clinical factors predicting postoperative complications to determine whether PsA-specific disease activity is associated with adverse events (AEs). Methods: We conducted a prospective study of adults with PsA undergoing THA or TKA. Baseline assessments included demographics, BMI, Charlson Comorbidity Index (CCI), PsA disease activity measures, patient global assessment, medications, and patient-reported outcomes. Operative synovial tissue was assessed for inflammation. Postoperative AEs collected within 1 year were graded for severity (1-4) using the Clavien-Dindo (CD) classification. Logistic regression, adjusting for age and sex, assessed predictors of AEs and AE severity. Results: In total, 57 patients were included (24 THA, 33 TKA). At surgery, patients had moderate PsA activity (mean Disease Activity Index for Psoriatic Arthritis [DAPSA] score of 15.8) with minimal skin disease or enthesitis. CD grade 3-4 AEs occurred in 10 patients (18%). BMI and CCI were higher in THA patients with any AEs and all patients with grade 3-4 AEs. In the adjusted analysis, higher CCI was associated with grade 2-4 and 3-4 AEs, and BMI with grade 3-4 AEs. No PsA-specific measures, including DAPSA or synovial histologic inflammation, were associated with AEs. Conclusion: In patients with PsA undergoing THA/TKA, AEs were driven by comorbidity burden and obesity rather than PsA-specific disease activity or joint inflammation. This finding highlights the importance of perioperative optimization of comorbidities and weight.

  • A Machine Learning Approach to Determine the Optimal Age for Total Hip Arthroplasty: When Is Risk for Adverse Outcomes Lowest?

    by Chloe Heiting, Yiyuan Wu, Susan M. Goodman, Peter Sculco, Fei Wang, Said Ibrahim, Peter Cram, Rich Caruana, Bella Mehta

    Introduction: With evolving lifestyles and improvements in surgical techniques, the utilization of total hip arthroplasty (THA) is growing across patients of all ages. Although complication rates are low, they are still a key consideration in THA planning. In this study, we explore the optimal ages for THA with the lowest risk of adverse outcomes. Methods: Patients who underwent primary elective THA were retrospectively identified from the Pennsylvania Health Care Cost Containment Council Database, 2012 to 2018. We trained a supervised model learning model—explainable boosting machines—with a 70% train to 30% test split to predict risk for 90-day readmission, 90-day mortality, 1-year revision, and longer length of stay (LOS). Using this “glass box” model, we assessed risk for these adverse outcomes as they change with age. We then visualized two-way interactions between age and other patient-level covariates in the models to further understand risks. Results: Our cohort included 105,336 patients with rates of 8.0% readmission, 0.3% mortality, and 1.5% revision. The median LOS was 2 days (interquartile range [1, 3]). We demonstrated a nonlinear relationship between age and THA outcomes, and identified age as one of the most important factors in prediction of adverse outcome. Risk for 90-day readmission increased between 48.5 and 71.5 years, and 90-day mortality risk increased after 79.5 years. Risk for 1-year revision increased before 46.5 years, and risk for longer LOS decreased between 52.5 and 74.5 years. Multiple factors influenced the relationship between age and THA outcomes, such as discharge location and comorbidities. Conclusions: Risk for adverse THA outcomes changes in a nonlinear manner with age. Our models suggest that the optimal age for lowest risk of adverse outcomes after THA is between 52.5 and 71.5 years. This study may provide nuanced risk quantification when considering age in THA, and inform planning and election of THA.

  • Open Access

    Exploring the optimal age for total knee arthroplasty to minimize risk of adverse outcomes: machine learning analysis of a statewide cohort

    by Chloe Heiting, Yiyuan Wu, Susan M. Goodman, Peter Sculco, Fei Wang, Said Ibrahim, Peter Cram, Rich Caruana, Bella Mehta

    Background: Rates of total knee arthroplasty (TKA) in the United States have risen in patients of a wide age range. Although rates of postoperative TKA complications have decreased, they remain a significant concern. In this study, we aim to determine how the risk of adverse TKA outcomes changes dynamically with age and explore the optimal ages with the lowest risk for adverse outcomes. Methods: This retrospective cohort study included patients who underwent elective primary TKA from 2012 to 2018 in the Pennsylvania Health Care Cost Containment Council Database. We trained (70% train:30% test) an explainable boosting machine (EBM), a modern generalized additive model, to predict risk for 90-day mortality, 90-day readmission, 1-year revision, and longer length of stay (LOS). This “glass box” model allowed us to measure and visualize feature importance using mean absolute scores and determine the role of age in the model. We then ran EBM models that allowed two-way interactions between age and patient-level covariates. Results: In our cohort of 227,959 patients, 90-day readmission was observed in 7.5%, 90-day mortality in 0.2%, and 1-year revision in 0.8%. The median LOS was 2 days (IQR [2, 3]). Age was among the most important factors for predicting all outcomes, and these were nonlinear relationships. The risk for 90-day mortality increased substantially at 76.5 years, and for 90-day readmission and longer LOS at 73.5 years. Risk for 1-year revision was greater before 63.5 years. Conclusions: We determined that there is a nonlinear relationship between age and risk for adverse TKA outcomes, and it changes dramatically at specific time points. Our data suggests that the optimal age for lower risk of 90-day mortality, 90-day readmission, and longer LOS is below 73.5 years, and above 63.5 years for 1-year revision. These findings can help in decision-making when trying to quantify risks related to aging.

  • Open Access

    Artificial intelligence in osteoarthritis research: summary of the 2025 OARSI pre-congress workshop

    by Matthew S. Harkey, Kerry E. Costello, Bella Mehta, Chunyi Wen, Anne-Marie Malfait, Henning Madry, Brooke Patterson


  • Open Access

    30-day readmission after total hip and knee arthroplasty in patients with systemic lupus erythematosus: A US nationwide study

    by Omkar S. Anaspure, Michael L. Parks, Jason Blevins, David Fernandez, Chisa Hidaka, Susan M. Goodman, Bella Mehta


  • Comparing Community‐Level Social Determinants of Health With Patient Race in Total Hip Arthroplasty Outcomes

    by Bella Mehta, Yi Yiyuan, Diyu Pearce‐Fisher, Kaylee Ho, Susan M. Goodman, Michael L. Parks, Fei Wang, Mark A. Fontana, Said Ibrahim, Peter Cram, et al.

    Objective: Social determinants of health (SDOH), including race, have a key role in total hip arthroplasty (THA) disparities. We compared the collective influence of community‐level SDOH to the influence of individual factors such as race, on THA outcomes.Methods: This retrospective cohort study of the Pennsylvania Health Care Cost Containment Council Database (2012–2018) included 105,336 patients undergoing unilateral primary elective THA. We extracted “community” factors from the US census by geocoding patient zip codes, including walkability index, household income, foreign‐born individuals, English proficiency, computer and internet access, unpaid family workers, those lacking health insurances, and education. We trained an explainable boosting machine, a modern form of generalized additive models, to predict 90‐day readmission, 90‐day mortality, one‐year revision, and length of stay (LOS). Mean absolute scores were aggregated to measure variable importance (ie, variables that contributed most to the prediction).Results: The rates of readmission, revision, and mortality were 8%, 1.5%, and 0.3%, respectively, with a median LOS of two days. Predictive performance measured by area under the receiver operating characteristic curve was 0.76 for mortality, 0.66 for readmission, and 0.57 for one‐year revision. For LOS, the root mean squared error was 0.41 (R2 = 0.2). The top three predictors of mortality were community, discharge location, and age; for readmission, they were discharge location, age, and comorbidities; for revision, they were community, discharge location, and comorbidities; and for LOS, they were discharge location, community, and comorbidities.Conclusion: Community‐level SDOH were significantly more important than individual race in contributing to the prediction of THA outcomes, especially for 90‐day mortality.


  • Open Access

    Patient‐Reported Fatigue Associated with Joint Histopathology in Rheumatoid Arthritis

    by Diyu Pearce‐Fisher, Melanie H. Smith, Bella Y. Mehta, Edoardo Spolaore, Edward DiCarlo, Dongmei Sun, Susan M. Goodman

    Objective: Fatigue is important for patients with rheumatoid arthritis (RA) but is poorly understood. We sought to study associations of fatigue with clinical features, disease activity, and synovial histology.Methods: Patients meeting the American College of Rheumatology/EULAR 1987 and/or 2010 RA criteria were recruited before elective total joint replacement. Demographics, RA characteristics, tender and swollen joints, erythrocyte sedimentation rate (ESR) and C‐reactive protein, and patient‐reported fatigue, categorized as mild, moderate, or severe, were collected. Hematoxylin and eosin stains of sectioned synovium were systematically scored by a pathologist. Relationships between fatigue and studied variables were evaluated with Kendall's tau. A directed acyclic graph (DAG) was used to illustrate associations of exposures, outcome variables, mediators, and confounders. Multivariable ordered logistic regression was used to further study associations.Results: Of 160 included patients, 85.6% were women, with a median age of 63.5 (55.25–71.40) and mean disease activity scores in 28 joints using ESR (DAS28‐ESR) of 3.91 (SD 1.3). There were no differences in comorbidities across fatigue categories. Fatigue correlated with DAS28‐ESR, synovial lining hyperplasia (SLH), anxiety, depression, and pain. In the DAG, DAS28‐ESR was associated with fatigue, full mediation by pain, partial mediation by depression and anxiety, and confounding by female sex. SLH was independently associated with fatigue but did not confound the relationship between DAS28‐ESR and fatigue. SLH was affected by synovial lymphocytic inflammation. In multivariable models, female sex, DAS28‐ESR, and SLH were all associated with higher fatigue.Conclusion: Although fatigue is associated with DAS28‐ESR, it is also associated with SLH independently of disease activity.