What does evidence mean in a world where cause and effect are often hidden – and how can we become better at interpreting what we observe?

It might help to read and re-read this very insightful and readable article by Sydney Brenner on meta-issues in biology
Sequences and consequences | Philosophical Transactions of the Royal Society B: Biological Sciences

Important quotations from the above article:
1.  "The conversion of data into knowledge constitutes a great challenge for future biological research."
2.  "It is known that inverse problems can only be solved under very specific conditions."
3.  "We need to remember that whereas mathematics is the art of the perfect and physics the art of the optimal, biology, because of evolution, is only the art of the satisfactory."
4.  "But, more importantly, it reveals the great principle that biological systems solve many problems by treating them like income tax. As is well known, it is criminal to evade income tax, but there are perfectly legal means of avoidance. Thus in this case, the problem of molecular complexity has not been directly solved, but avoided by the modular structure, which in turn simplifies it and also facilitates evolutionary change." 

Two key issues have to be kept in mind:

1) There will always be uncertainty and variability in every real-world system. You have to view everything through a stochastic lens and be very careful about letting simplifications, heuristics, and models lead you to assume the world is deterministic.

2) Bias can never be completely excluded from any measurement or theory. While the practice of science is designed to bring us as close to objective truth as humanly possible, no science (with the possible exception of some theoretical mathematics) can ever be truly 100% objective. Instruments will always have measurement error. Observations and theories will always have some degree of bias that comes from their originator. We cannot eliminate it entirely, so we must take care to use good scientific principles and methods to limit it to the extent possible, and then be cognizant that some bias will inevitably continue to exist.

The question is so tricky and is not limited to medicine only but it can include economy, politics, Sicology and others. 
Regarding medicine, for me and based on my personal experience e.g. in Anorectal physiology for example, where most of the factors which may be thought to be sure to lead to pathology such ad obesteric sphincter injury doesn't necessarily lead to Anal incontinence. In order to understand this finding as a cause and effect we have to understand the normal Anorectal physiology through a model using the flow equation. Using this model we could understand that Anorectal segment is a highly integrated system and an injury to the Anal sphincter has to be severe enough to make the Anal Canal resistance to be seriously low. Even in this situation if there is a rectal inertia this will have a protective effect to avoid Anal incontinence. 
In Economy , politics and sociology a serious and analytical review on the effect of any decision or event on the outcome e.g. the effect of political stability "irrespective to the political ideology" governed by ligeslative stability are attractive to investments, improved employment, financial income on more social stability.
At the end the cause and effect is from my personal point of view is best evaluated by Cross-over studies where all the confounding factors are eliminated.

While causality is difficult to establish from observational studies, it is possible to infer causality from such studies—namely, cohort and case-control studies—using a nuanced framework called the causality index. This index indicates whether an effect is causal or merely coincidental, and if causal, the extent of the effect. The framework is derived from three notable epidemiological models: the epidemiological triangle, Bradford Hill’s criteria, and Rothman’s causal pie model.

The derivation of this framework is grounded in the principle of data triangulation, meaning that more than a single study is required to deduce causality in an exposure–outcome relationship. By implication, causality cannot be established based on a single observational study. Efforts must be made to compare the findings of your study with those of previous studies that share similar characteristics. In fact, this is the primary purpose of the "Discussion" section in a manuscript or thesis.

The framework is applicable to case-control and cohort studies that carry a low to moderate risk of bias. The concept of causality can be misleading if the study in question is significantly flawed.

The causality domains include:

Regarding strength of association:

For consistency, calculate the ratio of statistically significant studies to the total number of studies that examined the exposure-outcome relationship. Causality cannot be determined based on a single study alone—you must compare your results with those of similar studies. In sampling previous studies, efforts must be made to avoid publication bias.

If your study is a cohort study, temporality is clearly established, and it is rated three. If it is a case-control study, it is rated two, but if the disease studied is rare, the rating is three.

Irreversibility, used as a proxy for biological gradient, is assessed by comparing the direction of your study's effect with that of previous studies. Tabulate your findings alongside previous studies with significant results, and examine the consistency in effect direction. Irreversibility is confirmed only if all effects point in the same direction (positive or negative), and it is rated one. If effects vary in direction (some positive, some negative), irreversibility is nullified and rated zero.

In total, the maximum causality score is 10. A higher score indicates a stronger likelihood of a causal relationship, with a score of 8 out of 10 typically used as the threshold for inferring causality.

 Relevant publications
1. Nweke M, Mshunqane N. Characterization and stratification of risk factors of stroke 

in people living with HIV: A theory-informed systematic review. BMC Cardiovascular 

Disorders. 2025;25(1):405. 

2. Nweke M, Oyirinnaya P, Nwoha P, Mitha SB, Mshunqane N, Govender N, Ukwuoma 

M, Ibeneme SC. Development of a high-performing, cost-effective and inclusive 

Afrocentric predictive model for stroke: a meta-analysis approach. BMC neurology. 

2025 Jul 7;25(1):282. 

3. Nwagha T, Nweke M. Stratification of risk factors of lung cancer-associated venous 

thromboembolism and determining the critical point for preemptive intervention: A 

systematic review with meta-analysis. Clinical Medicine Insights: Oncology. 2023 

Jun;17:11795549231175221.
4. Ibeneme SC, Odoh E, Martins N, Ibeneme GC. Developing an HIV-specific falls risk 

prediction model with a novel clinical index: a systematic review and meta-analysis 

method. BMC Infectious Diseases. 2024;24(1):1402. 

This is the wrong question to ask.

"How can we become better at ACTING based on what we observe?" 

Treating causality as binary implies the mantra "until we know everything, we shouldn't do anything."

Treating it as a continuum implies a more sensible dictum: "as we learn more, we should be willing to do more."

EVERY action-- including and especially doing nothing while awaiting more info-- has costs and benefits, and these costs and benefits are distributed in various ways among "winners and losers" who may have different characteristics (especially wealth).

The ONLY scientific and ethical way to act is to contemplate one or more actions, and compare it/them to the status quo, through the lens of good science (that is, "what do we know, with what uncertainty).

Too often, monied interests hide behind uncertainty.

 When causal relationships aren't immediately apparent, the evidence becomes especially complicated. This is actually the rule rather than the exception in the majority of important domains, such as economic systems, human behavior, and health outcomes. 
 
Evidence in these murky waters evolves from basic observation into a more complex endeavor involving probabilistic reasoning, controlled comparison, and pattern recognition. We are essentially acting as detectives when direct causation is concealed, assembling hints that might be dispersed by space, time, or layers of intervening circumstances.
 

Moving from "this happened, therefore that caused it" to "given what we observe, what are the most plausible explanations, and how can we test them?" is the crucial change. This necessitates recognizing the difference between correlation and causation as well as the fact that correlation can have meaning on its own.

Inferring causality in Biology from observational data alone is theoretically impossible; it is only by interventional (experimental, where we control the variables we want to see the causal effect on a response or set of responses) that we can establish causal relationships.
However, some statistical frameworks have been proposed (see doi: 10.1017/S0033291720005127)  to provide partial inference on causality in biological health related problems (see for example our paper  doi: 10.6026/97320630015372)

Re cause and effect: The effect will often be observable at least at a macroscopic level or at least some evidence based on smaller scales of observation. For example, many cancers or neurological disorders are not observable at a macroscopic level until late in the disease state, but microscopic/cellular and molecular changes can be observed with the appropriate analyses based on any suspiction that some molecular process is in progress. If none of these can be used, then it may be that there is no effect. In terms of cause, this is harder to determine and differential hypothesis that can be tested experimentally, or now perhaps with AI, may be available. Key here is the use of critical thinging that does several things if done properly: 1. Makes few assumptions about the cause a priori; 2. Is testable and in hypotheses that are falsifiable; 3. Is subject to revision with data. It is also crucial to keep in mind that cause may be multiple and perhaps interactive, may be time dependent as in progressive disease, and that apparently serendipidous results may actually be highly informative and should not a priori be discarded.

Evidence in the context of science and  where cause and effect are hidden means the research findings backed by authentic  research data and  results that were proven through critical thinking, well conducted clinical trials and thorough research findings that are proven and tested. 
The best way to interpret what we observe is by ensuring that research findings are properly  reported in scientific way using all the necessary statistical tools, graphics, tables, charts, and well articulated grammar that people will understand in both scientific snd non scientific communities