Signal-Not-Causation Study Check
Separate statistical signals from causal claims before repeating health headlines.
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 98%
The Signal-Not-Causation Check is a compact method for reading health research before accepting a headline. First identify whether the study was randomized, observational, retrospective, or merely a review of existing chart notes. Then inspect absolute event counts, comparison groups, time range, concurrent medications, pre-existing conditions, and other confounders. A detected association is labeled a signal unless the design and analysis justify a causal inference. The reader also checks the authors' own limitations and conclusion, which often contain cautions omitted from social media summaries. Finally, the claim being circulated is compared directly with the evidence. A legitimate signal can justify screening, monitoring, or further study without proving that a treatment caused an outcome. The framework converts emotional reactions into a repeatable evidence-checking sequence.
Origin
Extracted from Habits & Hustle, where Dr. Tyna Moore dissected chart-review claims connecting semaglutide with suicide and criticized causal headlines unsupported by the study.
Core principles
- 01Association does not establish causation.
- 02Study design determines which conclusions the evidence can support.
- 03Absolute event counts and comparison groups matter.
- 04Confounders can explain an apparent signal.
- 05Authors' stated limitations and conclusions should constrain interpretation.
How to run it
- 1
Classify the evidence
Identify whether the research is randomized, observational, retrospective, correlative, or another design. State what that design can and cannot establish.
Pro tip Look in the methods section rather than relying on the abstract or headline.
Watch out Large sample size does not convert observational data into a randomized experiment.
- 2
Inspect the actual numbers
Find the absolute number of events, group sizes, comparison rates, and study period. Separate statistical visibility from practical magnitude.
Pro tip Translate percentages into counts when possible.
Watch out Relative risk can sound dramatic when the underlying event is rare.
- 3
Search for confounders
Check concurrent drugs, previous diagnoses, disease severity, demographics, and selection effects that could influence both treatment and outcome.
Pro tip Ask what was true before the treatment began.
Watch out Chart data may omit important factors that were never recorded.
- 4
Read the limitations
Review the authors' explicit cautions, alternative explanations, and recommendations. Treat these as boundaries on responsible interpretation.
Pro tip Quote the conclusion directly when discussing the study publicly.
Watch out Do not promote a stronger claim than the researchers themselves make.
- 5
Label the conclusion correctly
Call an association a signal, correlation, or hypothesis unless causal evidence exists. Translate the signal into proportionate action such as screening or further research.
Pro tip Use language such as associated with rather than causes.
Watch out A signal should not be ignored, but it should not be sensationalized either.
In the wild
Moore described a review of millions of chart notes that found a small signal among patients who also used antidepressants and benzodiazepines. She emphasized that the authors listed numerous limitations and recommended mental-health screening rather than claiming semaglutide caused suicide.
→ The finding becomes a monitoring signal and research question, not a proven causal warning.
A small chart review finds that people taking both HRT and semaglutide lost more fat. The reviewer identifies the study as correlational, checks selection effects, and treats the result as interesting but insufficient to prescribe the combination for that purpose.
→ The claim remains appropriately tentative until stronger evidence is available.
Common mistakes
Turning correlation into causation
People happening to use a treatment when an outcome occurs does not prove that the treatment produced the outcome.
Ignoring absolute counts
A headline based on a tiny number of events can sound much more conclusive than the underlying data warrants.
Skipping the limitations section
Authors frequently state crucial uncertainties and confounders that disappear when a study is summarized online.
Is it for you?
Best for
Anyone evaluating health studies, safety scares, treatment claims, or research-based social media content.
Not ideal for
It is not a replacement for formal systematic review or expert statistical analysis when making high-stakes clinical policy.
From the transcript
“But again, it was correlative, not causative.”
“This is a signal. This is not causative.”
“And they even said in the conclusion that really what we found here is that we just need to be screening our patients for mental…”
From the episode
Episode 393: Dr. Tyna Moore: Why Ozempic Should Be Microdosed, HRT Tips, Peptides for Metabolic Health + More
Dr. Tyna Moore