In the current era, the precision medicine initiative is stressed as being a priority for the best medical care one can practice/get. In addition, a statistical method called bootstrapping with the support of strong AI (e.g. powerful processing, neural networks) and computers proves to be more reliable than 'classical statistics' used in biomedical millieu. One can generate statisticaly confident results without large samples, as long as the quantitavely smaller ones included are a representatively and correctly chosen values (with enough qualitative variations needed for the study). Some questions arise when considering these large shifts and new possibilites for generating and practicing medical knowledge.
1. Will large scale studies be redundant once precision medicine methods (such as (epi-)gene-orientated analysis for assessment of pharmacotherapy efficacy or enzyme status'), become more available and cheaper? It would surely prevent lots of heavier life-threatening side-effects (therefore costs, but also quality of patient's life) and a more favorable clinical outcome for the benefit of the patient (and the entire medical system).
2. Could these new 'large scale studies' be made through correctly chosen smaller samples run with bootstrapping method? (It would save time, money, generate equal or even more significant findings due to computing possibilites previously unavailable?)
3. an example; commonly used drugs that affect coagulation (warfarin, P2Y antagonist. new anticoagulants) can have terminal side effects that we want to evade. By having large scale studies from real population data we can assess the common risk factors/symptoms and derive scores for assessment (e.g. HAS-BLED or even the one's for primary prevention such as CHADS) and according to which we can expect direct consequences for patients but also project economic impact of such therapies. If one would use precision medicine, we could surely evade a lot of unwanted side effects that stem from drug-interactions or enzyme inhibitors/inducers concomitantly used, but also due to genetic alterations (e.g. coagulation cascade variability leading to a prothrombotic/prohemorrhagic state).
In what direction will/should the future of decision making process gravitate to? Statistics is something that never exists in reality, but could point an insecure clinician to a safer route, while precision medicine (getting more affordable with each moment passed), is a way that resepects the basic physiology and biology which is entirely unique in each individual. Will these two methods (small scale vs. large scale) find an equal purpose or are they irreconcilable in their nature?
https://www.nih.gov/precision-medicine-initiative-cohort-program
https://en.wikipedia.org/wiki/Bootstrapping_(statistics)
See my papers about "Matroska feature selection method for microarray data" from RG.
Dear George, I'm unable to open the link.
Can you please upload it here?
Regards,
Karlo
Speaking from experience, even large studies often fail to capture the complexity of the patient population. How would one know that a "carefully chosen representative sample" is in fact representative? In one study where we tried to predict patient satisfaction, we had data from 10,000 patients, which was a representative sample (as far as we could tell). We had thousands of predictor variables per patient. But no matter what AI or other predictive models we used, we could not achieve an AUC better than 0.55, i.e. not much better than a coin flip. I am skeptical of genetics papers that claim to have high predictive accuracy.
And yes, we tried stratifying the population into subgroups. But we discovered that any attempts at stratification resulted in overfitting and made the predictive accuracy worse. There are situations where predictions are intractable, at least based on the data available. Unless we develop remote scanning technology to read people's thoughts or something, predicting patient satisfaction in our population has been an impossible challenge thus far.
Thank you for contributing with this piece of first hand experience. It will truly be exciting (and I hope fruitful) to see what the future will bring and how will the current/future methods change our decision making in medical field, but more importantly the process of generating new knowledge. Biostatistics reading/wiritng should definitely count as basic medical litteracy, since shallow knowledge leads to clinicians being impressed by mathematics not fully comprehended. When reading even high-impact medical journals, it sometimes seems that the mere number of participants involved in a study covered by a lot of graphs produced by it gets the 'green light' for publishing, even though they havn't generated any useful or new findings. In the end, real patients depend on this interpretation that may end up as official guidelines or accepted practice.
- Badges
- Science topic
- Similar topics
- Chemistry
- Pharmacology
- Medical Pharmacology