Inhabitants-adjusted oblique comparisons (PAICs) embody each matching-adjusted oblique comparisons (MAICs) and Simulated remedy comparisons (STCs). The important thing knowledge requirement for these strategies is that they’ve particular person affected person knowledge (IPD) from at the very least one medical trial. This implies the strategies are most helpful for studied funded by the medical trial sponsor or when IPD medical trial are publicly out there (as an illustration, see the Yale College Open Knowledge Entry (YODA) Challenge.
Normal approaches to evaluating efficacy and security throughout medical trials when head-to-head trial knowledge shouldn’t be out there typically depend on oblique remedy comparisons (ITC, see Bucher et al. 1997) or community meta-analysis (NMA, see Dias et al. 2013). One key assumption for these strategies to be legitimate are that there are not any impact modifiers or the distribution of impact modifiers is similar throughout trials. This bias could also be notably massive when there are sparse networks of trials within the oblique comparability.
This can be a key benefit of PAIC, however there are additionally some disadvantages. A NICE DSU Technical Assist Doc (#18) from 2016 outlines a few of these benefits and limitations. On the constraints aspect:
Each MAIC and STC can be utilized to hold out both an “anchored” oblique comparability, the place there’s a frequent comparator arm in every trial, or an “unanchored” oblique comparability, the place there’s a disconnected remedy community or single-arm research. An unanchored MAIC or STC successfully assumes that absolute outcomes might be predicted from the covariates; that’s, it assumes that each one impact modifiers and prognostic components are accounted for. This assumption may be very sturdy, and largely thought-about not possible to satisfy.
Additional, whereas conceptually PAICs ought to produce sturdy estimates within the presence of impact modifiers, there’s restricted proof that accuracy materially improves. Lastly, through the use of PAIC to reweight data from a given medical trial, the evaluation inhabitants could also be kind of just like the goal inhabitants of an intervention.
The NICE DSU Technical report makes plenty of suggestions to practitioners:
- Anchored most popular. Use anchored comparisons until there’s an absence of a standard comparator.
- Reveal a necessity. PAIC ought to be used when one can present (i) that there are impact modifiers current and (ii) there are variations within the distribution of impact modifiers throughout trials.
- Impact modifiers. When utilizing MAIC, all impact modifiers ought to be adjusted for to make sure stability and scale back bias, however no purely prognostic variables. This method is beneficial since rising the variety of matching variables might inflate the usual error resulting from over-matching.
- Use linear scale. NICE recommends that oblique comparisons should be carried out on the standard linear predictor scale
- Figuring out a goal inhabitants. Usually, medical trials are carried out on a particular goal inhabitants. Utilizing PAICs reweight the trial pattern in a means that might differ from the goal inhabitants. Explicitly stating the goal inhabitants of curiosity is essential to confirm how the reweighting strikes the weighted pattern nearer or farther from the goal inhabitants.
- Clear reporting. Researchers ought to assess of covariate distributions, present proof that variables are impact modifiers, present the distribution of weights (if relevant), and calculate applicable measures of uncertainty.
For extra particulars, see the whole NICE DSU Technical Assist Doc (#18) .