CORE Seminar - Dr. Nandini Dendukuri - September 8, 2025

Value of the Information & Strength of the Argument

The presentation provides valuable insights into a puzzling phenomenon in diagnostic test accuracy research. The use of simulation studies to isolate individual biases is a strong methodological approach, allowing for clear causal inference. The argumentation is logical and well-supported by examples from real meta-analyses, such as the GeneXpert TB test. The speaker effectively demonstrates how biases like imperfect reference standards can create spurious associations between prevalence and test accuracy, challenging the assumption of independence. The discussion of DAGs, while acknowledging their learning curve, offers a structured framework for understanding these biases. The talk is particularly valuable for researchers in clinical epidemiology and biostatistics, as it highlights the need for careful consideration of potential biases in meta-analyses of diagnostic tests.

Scientific Rigor, Source Quality, Title Accuracy

The presentation demonstrates high scientific rigor, with a clear theoretical framework and simulation-based evidence. The speaker cites relevant literature, including a 2009 paper by Leeflang et al. and a 2023 study in CJ, to support the existence of the phenomenon. The use of DAGs is methodologically sound, though the speaker admits to initial difficulty in interpreting them. The title accurately reflects the content, and the presentation stays focused on the research question. No external sources are provided in the description, but the speaker references her own work and collaborations. The audience interaction shows engagement and clarification of concepts, indicating a well-received presentation.

The title accurately reflects the content, which explores whether observed associations between prevalence and test accuracy are due to bias or biological variation.

8/10

The presentation is based on rigorous methodological research, including simulation studies and DAGs, presented by an expert in biostatistics. The content is well-structured and grounded in established epidemiological concepts. However, it is a seminar presentation without peer-reviewed publication details, and some claims rely on the author's interpretation of DAGs.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction by Benjamin Smith and Paulo Malo, followed by Dr. Dendukuri's opening remarks.
• 03:00Definition of prevalence, sensitivity, specificity, and predictive values using a 2x2 table.
• 07:00Example of GeneXpert TB test meta-analysis showing apparent association between specificity and prevalence.
• 10:00Second example of lymph node TB showing sensitivity increasing with prevalence, attributed to imperfect reference standard.
• 15:00Discussion of Leeflang et al. 2009 and a 2023 CJ study showing varying patterns of prevalence-accuracy associations.
• 20:00Introduction to DAGs and their application to diagnostic test accuracy studies.
• 25:00Simulation results showing that imperfect reference standards create artificial associations between prevalence and sensitivity/specificity.
• 30:00Simulation of conditional dependence and spectrum effects, further demonstrating bias-induced associations.
• 35:00Discussion of implications and future research directions.

• Leeflang et al. 2009 (Journal of Clinical Epidemiology) — Referenced as a paper that reported diagnostic test accuracy may vary with prevalence.
• CJ 2023 study (Journal of Clinical Epidemiology) — Referenced as a study analyzing 6,999 diagnostic test accuracy studies, finding sensitivity increasing and specificity decreasing with prevalence.

• Leeflang et al. 2009 (Journal of Clinical Epidemiology) — Supports the observation that test accuracy may vary with prevalence.
• CJ 2023 study (Journal of Clinical Epidemiology) — Provides empirical evidence of prevalence-accuracy associations across many studies.

• Theoretical independence of sensitivity and specificity from prevalence — The theoretical definition of sensitivity and specificity implies they should not vary with prevalence, contrasting with empirical observations.