This is something I posted on Twitter last October. Extended discussion can be found below the tweets
Propensity score matching is often used when two cohorts of patient are compared. Aim is to have somewhat comparable groups so effect of treatment or intervention could be estimated reliably. In short, two groups of patient tread by methods A and B are collected. Then subgroups of patients are selected based on some matching variables or covariates like age and BMI. My take is that no trust should be placed on studies using propensity scoring comparing operative and nonoperative treatment of orthopaedic or traumatological conditions because residual confounding is huge.
Basic concept on shared decision making is to recognize patient values, expectations and preferences when choosing between treatment options. These concepts may be very influential and have large effect on outcome. Patient may be catastrophizing or mentally fragile. In such cases it is not probably wise to choose operative treatment and take additional risks. Most importantly the treatment decision in this case is based on qualitative measure, not on quantitative covariate such as age, BMI, disease grade or smoking status. Another patient may expect a very fast return to work and is willing to take risks associated to surgery. It would not be wise to “force” patient to nonoperative treatment. Subjective tolerance for poorer outcome would be very low. While SDM is somewhat new concept it is evident that we have behaved in such way as long as we have acknowledged that absolute superiority does not exists between available treatment options. Usually this is the case in situations like clavicle fracture, achilles tendon rupture or various degenerative disease where we can measure numerous equal outcomes which all are value-laden.
These qualitative factors presumably have huge how impact when choosing between different treatment regimes. Comparing to common variables used to match patients in propensity scoring, like age, sex, smoking and BMI, these latent factors are very hard to quantitatively measure and use in propensity scoring. Hence I strongly think that propensity score method is next to useless in orthopaedics due to residual and uncontrollable confounding resulting from these poorly accountable latent factors.