Featured Pieces & Posts
Statistical evidence isn’t the problem. The problem is the simple alchemy of transmuting “statistically significant” into (scientifically) “significant” by dropping one word: A scientific falsehood from a small but significant change.
We must continue to assess the statistical quality of our study’s evidence (using p-values, confidence intervals, credible intervals, etc.) for our scientific finding. This is good for science.
We must stop confusing the statistical quality of our study’s evidence for our scientific finding (“statistical significance”) with the scientific quality of our study’s finding (“significance”). We do this every time we shorten “statistically significant” to “significant”. This is bad for science.
There was a significant decrease of D in the outcome.
There was no significant association between variables X and Y.
Ask yourself if a randomized controlled trial’s reported effect size estimate is meaningful, regardless of sample size.
Train yourself to internalize that significance does not imply importance.
Remember that sample size does not correlate with effect size.
Never just say “significant” when you really mean “statistically significant”. You will be misunderstood as saying “important”. Instead, always say or write out the whole phrase “statistically significant”.
We would overstate our health app’s effectiveness by claiming it reduces the risk of new coronavirus infections by 16.9% — when in fact it will only reduce this risk by 3.1%.
But we can re-weight our real-world evidence results to provide more accurate risk-reduction estimates of either 2.3% or 2.2%.
We would overstate our telemedicine app’s effectiveness by claiming it reduces the risk of new coronavirus infections by 16.9% — when in fact it will only reduce this risk by 3.1%.
EJ Daza. Towards Data Science.
This is exactly the time to temper the sprinting agility of data science with the scientifically rigorous methodology of biostatistics.
EJ Daza, K Wac, M Oppezzo. Healthcare.
Sleep deprivation is a prevalent and rising health concern, one with known effects on blood glucose (BG) levels, mood, and calorie consumption. However, the mechanisms by which sleep deprivation affects calorie consumption (e.g., measured via self-reported types craved food) are unclear, and may be highly idiographic (i.e., individual specific). Single-case or “n-of-1” randomized trials (N1RT) are useful in exploring such effects by exposing each subject to both sleep deprivation and baseline conditions, thereby characterizing effects specific to that individual. We had two objectives: (1) To test and generate individual-specific N1RT hypotheses of the effects of sleep deprivation on next-day BG level, mood, and food cravings in two non-diabetic individuals; (2) To refine and guide a future n-of-1 study design for testing and generating such idiographic hypotheses for personalized management of sleep behavior in particular, and for chronic health conditions more broadly. We initially did not find evidence for an idiographic effect of sleep deprivation, but better-refined post hoc findings indicate that sleep deprivation may have increased BG fluctuations, cravings, and negative emotions. We also introduce an application of mixed-effects models and pancit plots to assess idiographic effects over time.
Jan 2019 (In Preparation)
EJ Daza. arXiv.
Single-subject health data are becoming increasingly available thanks to advances in self-tracking technology (e.g., mobile devices, apps, sensors, implants). Many users and health caregivers seek to use such observational time series data to recommend changing health practices in order to achieve desired health outcomes. However, there are few available causal inference approaches that are flexible enough to analyze such idiographic data. We develop a recently introduced framework, and implement a flexible random-forests g-formula approach to estimating a recurring individualized effect called the "average period treatment effect". In the process, we argue that our approach essentially resembles that of a longitudinal study by partitioning a single time series into periods taking on binary treatment levels. We analyze six years of the author's own self-tracked physical activity and weight data to demonstrate our approach, and compare the results of our analysis to one that does not properly account for confounding.
EJ Daza. Methods of Information in Medicine.
Conclusions. Causal analysis of an individual's time series data can be facilitated by an n-of-1 randomized trial counterfactual framework. However, for inference to be valid, the veracity of certain key assumptions must be assessed critically, and the hypothesized causal models must be interpretable and meaningful.
ML Nguyen, J Hu, K Hastings, E Daza, M Cullen, L Orloff, L Palaniappan. Cancer.
Conclusions. Negative prognostic factors for thyroid cancer traditionally include age >45 years and male sex. The results of the current study demonstrate that Filipinos die of thyroid cancer at higher rates than NFA and NHW individuals of similar ages. Highly educated Filipinos and Filipino women may be especially at risk of poor thyroid cancer outcomes. Filipino ethnicity should be factored into clinical decision making in the management of patients with thyroid cancer.
EJ Daza, MG Hudgens, AH Herring. The Stata Journal.
Individuals may drop out of a longitudinal study, rendering their outcomes unobserved but still well defined. However, they may also undergo truncation (for example, death), beyond which their outcomes are no longer meaningful. Kurland and Heagerty (2005, Biostatistics 6: 241–258) developed a method to conduct regression conditioning on nontruncation, that is, regression conditioning on continuation (RCC), for longitudinal outcomes that are monotonically missing at random (for example, because of dropout). This method first estimates the probability of dropout among continuing individuals to construct inverse-probability weights (IPWs), then fits generalized estimating equations (GEE) with these IPWs. In this article, we present the xtrccipw command, which can both estimate the IPWs required by RCC and then use these IPWs in a GEE estimator by calling the glm command from within xtrccipw. In the absence of truncation, the xtrccipwcommand can also be used to run a weighted GEE analysis. We demonstrate the xtrccipw command by analyzing an example dataset and the original Kurland and Heagerty (2005) data. We also use xtrccipw to illustrate some empirical properties of RCC through a simulation study.
AP Keil, EJ Daza, SM Engel, JP Buckley, JK Edwards. Statistical Methods in Medical Research.
Epidemiologists often wish to estimate quantities that are easy to communicate and correspond to the results of realistic public health interventions. Methods from causal inference can answer these questions. We adopt the language of potential outcomes under Rubin’s original Bayesian framework and show that the parametric g-formula is easily amenable to a Bayesian approach. We show that the frequentist properties of the Bayesian g-formula suggest it improves the accuracy of estimates of causal effects in small samples or when data are sparse. We demonstrate an approach to estimate the effect of environmental tobacco smoke on body mass index among children aged 4–9 years who were enrolled in a longitudinal birth cohort in New York, USA. We provide an algorithm and supply SAS and Stan code that can be adopted to implement this computational approach more generally.
CG Brown-Johnson, A Burbank, EJ Daza, A Wassmann, A Chieng, GW Rutledge, JJ Prochaska. American Journal of Preventive Medicine.
Conclusions. Examination of online patient–provider communications provides insight into consumer health experience with emerging alternative tobacco products. Patient concerns largely related to harms and safety, and patients preferred provider responses positively inclined toward e-cigarettes. Lacking conclusive evidence of e-cigarette safety or efficacy, healthcare providers encouraged smoking cessation and recommended first-line cessation treatment approaches.
JJ Prochaska, AK Michalek, C Brown-Johnson, EJ Daza, M Baiocchi, N Anzai, A Rogers, M Grigg, A Chieng. JAMA Internal Medicine.
Conclusions and Relevance. To our knowledge, this is the first study to prospectively track reemployment success by smoking status. Smokers had a lower likelihood of reemployment at 1 year and were paid significantly less than nonsmokers when reemployed. Treatment of tobacco use in unemployment service settings is worth testing for increasing reemployment success and financial well-being.