Raj Mehta Profile picture
Family Doc, Clinical Informaticist, EBM & Bioethics enthusiast | Faculty @AdventHealth Family Medicine Residency | @UFMEDICINE alum
Jul 29, 2022 15 tweets 5 min read
1/ Thread on Analyzing Physician Thought Processes with Conditional Probability

One important aspect of #EBM is to help physicians calibrate their probabilistic reasoning and improve their clinical judgement.

We will walk through one example from study highlighted below. 2/ The Study:

The basis of this thread is this interesting paper on physician probability estimates of medical outcomes:

jamanetwork.com/journals/jaman…
Sep 13, 2021 14 tweets 5 min read
1/ Tweetorial on Stratification bias (or Collider-Stratification bias)

This is something that comes up often in Obs studies. To quote @Lester_Domes:

"It has long been known that stratifying on variables affected by the study exposure can create selection bias" 2/ Selection bias is a common problem in Observational Research.

Improper stratification can inadvertently cause selection bias. The paper below is an excellent introduction to this topic.

journals.lww.com/epidem/Fulltex…
Apr 18, 2021 24 tweets 5 min read
1/ A thread on discussing airborne transmission of respiratory diseases

or

How the droplet vs airborne debate is deceptively about not one, not two, but *three* different dichotomies.

(warning: this thread is really long. sorry!) 2/ This thread was inspired in part from great discussions after the tweet below.

Many agree that dichotomizing the droplet/aerosol is wrong, but ignore dichotomizing of:

-> infectiousness of respiratory pathogens
&
-> effect size of infection control

Apr 16, 2021 14 tweets 4 min read
1/ A brief thread on how I think about diagnostic reasoning & measurement error

Or

How regression to the mean and confounding bias leads docs to over-estimate both the probability of disease & effectiveness of interventions.

2/
True findings - absence or presence of abnormalities *due to disease* that can effect outcomes.

Observed findings - signs, symptoms, lab results, radiology, etc.

Diagnosis/Intervention - based on observed findings

Outcomes - effected by true findings & interventions Image
Mar 11, 2021 5 tweets 2 min read
1/ The success of the UK Recovery trials both inspires and frustrates.

Inspirational in it's achievement, but frustrating when I compare it to efforts the US.

Why has the US has not been able to replicate similar research for therapeutic agents?

2/ Convalescent plasma (CP) is the best example of the all too common US failure.

First a question. The UK Recovery Trial had to enroll 11k patients before they found futility.

Meanwhile, how many hospitalized patients in the US received CP without being enrolled in any RCTs?
Jun 24, 2020 22 tweets 6 min read
1/ Tweetorial at use of Race/Ethnicity in prediction models, specifically looking at eGFR based on CKD-EPI formula.

eGFR = estimated GFR

ckdepi.org 2/ First some background.

GFR can be directly measured, traditionally by measuring iothalamate clearance. This is process is time consuming, costly, and rather inconvenient. So for greater ease, we attempt to estimate GFR (eGFR) based on serum measurements of Creatinine.
May 24, 2020 15 tweets 4 min read
1/ Tweetorial on Decision making and Evidence Based Medicine

Or

Integrating Bayesian Inference, Utility, and Uncertainty to optimize rational Medical Decisions. 2/ General disclaimer, this is more of a proposal on how to *add* decision making into our existing EBM framework.

Feedback is welcome, and this is hopefully an evolving field thanks to @RichardLehman1 and others!
May 22, 2020 15 tweets 5 min read
1/ Thread on evaluating benefits vs harms of masks for infectious diseases & #covid19.

or

Why i think the claim that masks have equal probability of causing harm & benefit is mistaken. 2/ General notice, I am not an epidemiologist!

This thread was inspired partly by conversations with @TheSGEM, and how clinicians approach balancing benefits/harms for various interventions.
May 4, 2020 15 tweets 5 min read
1/ Tweetorial on two common sources of bias in Observational Studies:

Confounding by Indication
Immortal Time Bias

This a continuation of Tweetorials on Causal Inference in EBM. For an introduction to the topic, start here:

2/ "Confounding by Indication"

Let us say we have a cohort study looking at the effects of ACE-I on death.

We define our cohort based on treatment assignment (did they get/not-get the RX).

Exposure: Rx Cohort (green)
Outcome: Death (blue) Image
May 1, 2020 13 tweets 4 min read
1/ Brief explanation *why* researchers for the NIAID trial on Remdesivir decided to change the primary outcome.

Essentially, as the trial extended beyond 14 days, "time to event" become a superior measurement choice.

h/t @AdanZBecerra1

2/ In study design, more study Power means we are more likely to notice a treatment effect (assuming an effect exists).

One way to improve study Power is to increase the sample size.

Another way is to improve our choice of measurement
Apr 27, 2020 14 tweets 5 min read
1/ Tweetorial on this WSJ article, and the importance of understanding confounders & colliders when making causal inferences.

The article asks a good question, but unfortunately makes several errors. Lets take a look!

wsj.com/articles/do-lo… 2/ The articles asks "Do quick shutdowns work to fight the spread of COVID-19?". A good causal question!

Here is a simple DAG to represent the inference:

Intervention: Shutdown
Outcome: Total COVID19 Deaths

The "arrow" is the effect of the intervention on the outcome.
Apr 10, 2020 22 tweets 6 min read
1/ Tweetorial on Cause & Effect in Evidence Based Medicine

Alternatively: why we need to know the natural history of disease

Causal reasoning is a core part of the practice of medicine, but it is sometimes neglected in our EBM curriculum.

So here is a brief introduction! 2/ The main references are wikipedia, and the Causal Inference Handbook by Hernan & Robins

hsph.harvard.edu/miguel-hernan/…
Apr 10, 2020 10 tweets 3 min read
1/ List of EBM tweetorials!

False Positives & Transposing the conditional:

2/

Using the MCID to evaluate Clinical vs Statistical significance