About

Anil Vachani, MD, MSCE, is an Emeritus Professor of Medicine in the Pulmonary, Allergy, and Critical Care Division at the University of Pennsylvania Perelman School of Medicine. He serves as the Chief of Medicine (Pulmonary, Allergy, and Critical Care) and is an Attending Physician at the Philadelphia Veterans Administration Medical Center. His roles include Director of Clinical Research in the Section of Interventional Pulmonology and Thoracic Oncology, as well as Director of Bronchoscopy at the Philadelphia VA Medical Center. Additionally, he is the Co-Director of the Lung Cancer Screening Program at both the Perelman School of Medicine and the Philadelphia VA Medical Center.

Research topics

• Internal medicine
• Computational biology
• Pathology
• Intensive care medicine
• Biology
• Bioinformatics
• Genetics
• Family medicine
• Radiology
• Medicine

Selected publications

• Impact of non-tumor size related features on staging in non-small cell lung cancer: a population-level analysis of the SEER-medicare database

  Journal of Thoracic Disease · 2026-03-01

  articleOpen access

  Abstract: The objective of this study was to quantify and characterize the presence of non-tumor size features among non-small cell lung cancer (NSCLC) patients in a surgical cohort versus those in a non-surgical cohort and assess the potential impact of key features on staging. A total of 22,876 patients aged 66 years and older with newly diagnosed (2010–2015) stage I–III NSCLC (T1–T3, N0/N1, M0) without prior or concurrent primary cancer were identified from the Surveillance, Epidemiology, and End Results (SEER)-Medicare Database. Reported T stage data (American Joint Committee on Cancer 7th Edition) were compared with a hypothetical tumor size only T stage (Tsize-Only), which did not account for non-tumor size defining characteristics. Frequencies of key non-tumor size features were analyzed among surgical and non-surgical cohorts. In the surgical cohort, a lower Tsize-Only vs. SEER-reported T stage was observed in 2,531 patients (20.8%). Specifically, T stage would be lower in 1,542 (32.6%) patients with T2 disease and in 989 (66.4%) patients with T3 disease. In the non-surgical cohort, lower Tsize-Only vs. Reported T was observed in 1,446 patients (13.5%) [342 (8.8%) patients with T2 disease and 1,104 (60.1%) patients with T3 disease]. The largest disparity between cohorts was in T2 disease and visceral pleural invasion (VPI): VPI was identified in 87.5% of surgically resected patients with lower Tsize-Only versus only 37.4% in non-surgical patients. Our findings highlight the need for better T stage characterization in non-surgical patients. Development of predictive models or standardized radiographic methods to accurately identify VPI in the non-surgical NSCLC population are recommended in light of the established impact of VPI on clinical outcomes. Trial investigators should consider the potential impact of un-identified VPI on their effect size estimates and sample size calculations. Attention towards non-tumor size related features, particularly VPI, may lead to improved treatment decisions and outcomes

  PublisherDOI

• Impact of Race and Acculturation on Lung Cancer Screening Eligibility: A Multi-Center Lung Cancer Cohort Study

  American Journal of Respiratory and Critical Care Medicine · 2026-03-19

  article
  PublisherDOI

• Monitoring the Harms of Lung Cancer Screening: Why the Nonmalignant Resection Rate Is Our Best Bet

  Journal of the American College of Radiology · 2025-12-01 · 3 citations

  articleOpen access
  PublisherOA PDFDOI

• Supplementary Figure 3 from Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care

  2025-11-26

  articleOpen access

  <p>Probability Threshold Plot. The figure displays the probability threshold plot for the complete range of the model's operating characteristics.</p>

  PublisherDOI

• Supplementary Table 3 from Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care

  2025-11-26

  articleOpen access

  <p>Metrics related to model utility in temporal validation in Optum EHR</p>

  PublisherDOI

• Supplementary Table 2 from Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care

  2025-11-26

  articleOpen access

  <p>List of 278 predictors and coefficient values in the final EHR model</p>

  PublisherDOI

• Supplementary Table from Evaluating and Improving Cancer Screening Process Quality in a Multilevel Context: The PROSPR II Consortium Design and Research Agenda

  2025-11-26

  articleOpen access

  Supplementary Table from Evaluating and Improving Cancer Screening Process Quality in a Multilevel Context: The PROSPR II Consortium Design and Research Agenda

  PublisherDOI

• 99P Radiologists perception on AI/ML software as a medical device (SaMD) unveiled via post-study usability survey: Key assets to redefine lung cancer screening practice

  ESMO Real World Data and Digital Oncology · 2025-11-01

  articleOpen access

  Background: Accurate prognosis in mNSCLC treated with immunotherapy can guide clinical decisions.We developed machine learning (ML) models to predict overall survival (OS) at 6, 12, 18, and 24 months in mNSCLC patients receiving pembrolizumab, based on clinicopathological features. Methods:We retrospectively analysed mNSCLC patients treated with pembrolizumab at the Prof.Dr. Ion Chiricuta Oncology Institute in Cluj-Napoca, Romania, from 2018 to 2023.Patients with missing data or who received <4 cycles of pembrolizumab were excluded, yielding 124 patients.We recorded clinical and paraclinical variables, including age, Eastern Cooperative Oncology Group performance status (ECOG PS), primary tumour T stage, presence of bone or brain metastases, multiple metastatic sites, prior palliative radiotherapy, neutrophil-to-lymphocyte ratio, lactate dehydrogenase, platelet-to-lymphocyte ratio, systemic immuneinflammation index (SII), and PD-L1 status.To address class imbalance, we applied SMOTE.Two SVM models were then trained with an 80/20 train-test split using 5fold cross-validation: a standard SVM and a weighted SVM in which each feature was weighted according to its hazard ratio from a multivariate Cox regression on OS.Results: Model A achieved accuracies of 90%, 82.17%, 82.30%, and 68.50% with areas under the curve (AUC)s of 0.39, 0.84, 0.87, and 0.69 for 6-, 12-, 18-, and 24months OS.Model B reached accuracies of 89.53%, 80.53%, 83.1%, and 69.3% and AUCs of 0.60, 0.86, 0.85, and 0.72 at the same timestamps.SHAP identified ECOG PS, age, multiple metastatic sites, NLR and SII as strongest predictors. Conclusions:These findings indicate that SVM-based algorithms integrating routine clinical data and inflammatory biomarkers can effectively predict survival in mNSCLC patients treated with pembrolizumab.Incorporating Cox-derived prognostic weights for each feature improved model accuracy, suggesting that combining traditional survival analysis with ML enhances discriminative performance.Such AI-driven prognostic tools may help stratify patients by risk and personalise treatment decisions.Validation in larger, independent cohorts is needed to confirm generalisability.

  PublisherOA PDFDOI

• Supplementary Table from Evaluating and Improving Cancer Screening Process Quality in a Multilevel Context: The PROSPR II Consortium Design and Research Agenda

  2025-11-26

  articleOpen access

  Supplementary Table from Evaluating and Improving Cancer Screening Process Quality in a Multilevel Context: The PROSPR II Consortium Design and Research Agenda

  PublisherDOI

• Supplementary Figure 1 from Machine Learning and Real-World Data to Predict Lung Cancer Risk in Routine Care

  2025-11-26

  articleOpen access

  &lt;p&gt;Flow Chart of Cohort Attrition and Study Design. A, shows the cohort attrition after applying eligibility criteria and the distribution of samples used for training and test sets. B, shows the parameters used for model development and study design.&lt;/p&gt;

  PublisherDOI

Recent grants

• Admin_Core

  NIH · $9.0M · 2022–2027

• NIH Grant R21CA156087

  NIH · $414k · 2014

• Environmental Health Informatics Core

  NIH · $16.5M · 2006–2030

• NIH Grant K07CA111952

  NIH · $692k · 2012

• Center for Research to Optimize Precision Lung Cancer Screening in Diverse Populations

  NIH · $12.8M · 2018–2025

Frequent coauthors

• Katharine A. Rendle

  Abramson Cancer Center

  101 shared

• Steven Μ. Albelda

  83 shared

• Jeffrey C. Thompson

  University of Pennsylvania

  62 shared

• Corey J. Langer

  University of Pennsylvania

  60 shared

• William N. Rom

  54 shared

• Louise C. Showe

  54 shared

• Michael K. Showe

  53 shared

• Justin E. Bekelman

  University of Pennsylvania

  49 shared