biostatistics
March 5, 2024
Statistical Fragility in Surveys: A Cautionary Tale
A recent study by Carnes et al. aimed to examine the effect of a workshop on overcoming bias and improving work climate in departments of medicine. While the goals were admirable, the results should be interpreted with caution due to statistical fragility. Low and uneven response rates, small effect sizes, and multiple significance tests raise…
January 3, 2024
Quantifying Unpredictability: Exploring Applications of the Heston Model in Healthcare
A financial model used to quantify unpredictable fluctuations in the stock market shows promise for improving analysis in healthcare. The Heston Model captures volatile dynamics like sudden surges and gradual reversion to the mean. This could translate to better epidemiologic forecasting and optimization of medication dosing. Rigorous testing is needed, but the versatility of mathematical…
November 29, 2023
The Ethical Significance of Statistics in Medical Research
A new review article examines statistics’s pivotal role in upholding research integrity and patient safety. It scrutinizes common statistical errors in medical studies, like p-hacking and neglecting limitations. Using historical examples, it demonstrates how such errors have resulted in flawed conclusions causing public harm, including hormone therapy research. Recommendations encompass more statistical rigor by ethical…
November 27, 2023
Statistical Significance Versus Clinical Relevance
This study investigated how often statistically significant results in nuclear medicine clinical trials actually translate into clinically meaningful differences for individual patients. The authors analyzed 32 test results reported as statistically significant and found that the cutoff between normal and abnormal values averaged just 0.66 standard deviations from the mean. This means that for a…
November 3, 2023
Comparing Statistical Significance with Clinical Relevance: The Fragility Index vs. The Relative Risk Index
A recent study compared different metrics for evaluating the fragility and clinical relevance of research findings. The fragility index, which measures how easily p-values flip from significant to nonsignificant, strongly correlated with p-values in simulated 2×2 contingency tables. This suggests the fragility index provides minimal insight beyond p-values alone. In contrast, the relative risk index,…
September 26, 2023
Normalizing Predictive Values for Appropriate Interpretation of Diagnostic Test Results
As clinicians, we often evaluate diagnostic test results like imaging studies to guide patient care. An important consideration is how disease prevalence in the tested population impacts the meaning of positive and negative predictive values. For example, a recent study using cardiac CT angiography (CTA) to diagnose coronary artery disease reported an excellent 99% negative…
September 5, 2023
The Value of Standardized Predictive Values in Diagnostic Testing
This letter to the editor published in the Journal of Magnetic Resonance Imaging discusses the usefulness of standardizing the predictive value of diagnostic tests. The author, Dr. Thomas F. Heston, argues that presenting predictive values standardized to a disease prevalence of 50% is more clinically useful than using the Predictive Summary Index. He explains that…
August 31, 2023
Going Beyond P-Values: A New Statistical Tool to Evaluate Research Findings
A recent study proposes an innovative statistical tool called the Robustness Index to quantify research findings’ fragility. As concern grows over issues reproducing published research, this new metric aims to provide a simple, interpretable measure of a study’s robustness against violations of assumptions. By examining how changing sample size affects significance, the Robustness Index adjusts…
August 22, 2023
The Predictive Power of Statistical Significance Depends on Study Power
A common misconception is that statistical significance is defined simply as a P-value of 0.05 or less. However, this definition does not account for study power. As this article explains, statistical significance should be determined using the positive predictive value, which factors the P-value and the power. Lower-powered studies require lower P-values to achieve statistical…
August 20, 2023
Improving the Clinical Utility of MRI Research
Predictive values like negative predictive value depend strongly on disease prevalence. To improve comparability between studies, authors should report standardized predictive values calculated at 50% disease prevalence. This allows readers to better judge clinical utility without performing further calculations. Standardization reduces prevalence bias when comparing diagnostic tests. Heston TF. Standardizing predictive values in diagnostic imaging…
Next page