Read on Twitter
The new Psychotherapy Research Special Issue on Machine Learning begins with a primer by Jaime Delgadillo @iapt_prn on machine learning for psychotherapy researchers: https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1859638 https://twitter.com/DepressionLab/status/1344235113979150336
Next up is a nice paper by @lluaces and colleagues in which they construct a prognostic index in the TADS study. Funnily enough, this is one of three simultaneously analyzed papers in which the TADS data were used for predictive modeling. https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1747657
The other two that used the TADS data similarly are Bondar et al., 2020 and Foster et al., 2019:
https://www.sciencedirect.com/science/article/pii/S2215036620300602 https://www.sciencedirect.com/science/article/pii/S0022395618302383
https://www.sciencedirect.com/science/article/pii/S2215036620300602 https://www.sciencedirect.com/science/article/pii/S0022395618302383
Next up is a paper by Lorimer and colleagues dynamically predicting risk of relapse over the 12 month period following completion of low intensity CBT https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1733127
(a surprising/depressing note to this study is that 70% of the individuals who were followed relapsed)
(a surprising/depressing note to this study is that 70% of the individuals who were followed relapsed)
Then we have a paper by the wonderful @Schwartz_PsyRes et al. in which we developed a treatment selection model in a naturalistic sample (N=966) treated w CBT or PDT and then evaluated the model in a true holdout (N=412) https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1769219
Next up Hilbert and colleagues make prognostic predictions in a sample of individuals with OCD undergoing CBT. https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1839140
Next, Kilcullen and colleagues look at data from a large practice research network of university counseling centers and identify predictors associated with individuals who return to therapy: https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1762948
In 1 of my favorite papers of 2020, the amazing MD PhD @SvanBronswijk (who just successfully defended) used two RCTs of CBT vs IPT to construct & externally validate treatment selection models; this is 1 of the 1st tests of generalizability in this area! https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1823029
in the external validation, Double Doctor van Bronswijk and colleagues observed (as we've long feared) significant shrinkage, and found no significant differences between those who received their indicated and non-indicated txs. many possible contributing factors to explore...
this study is noteworthy for our subfield bc it emphasizes the importance of external validation. psychotherapy RCTs are rarely large enough to support this sort of work, so we'll have to turn to other methods to get larger Ns (eg individual-patient-data meta-analytic approaches)
next we have a nice scoping review by Aafjes-van Doorn et al looking at how machine learning is used in psychotherapy research: https://www.tandfonline.com/doi/full/10.1080/10503307.2020.1808729 this reminds me of several similar recent pubs in other areas, all of which highlight widespread issues we need to address
e.g., this review looking at prediction models in psychiatry: Implementing Precision Psychiatry: A Systematic Review of Individualized Prediction Models for Clinical Practice https://academic.oup.com/schizophreniabulletin/advance-article/doi/10.1093/schbul/sbaa120/5903901#.X-xmfHHLGHs.twitter punchline: lots of papers that fail to clear some very basic, low bars.
but don't worry, it's not a problem unique to psychology/psychiatry: https://link.springer.com/article/10.1186/1471-2288-14-40 prediction modeling efforts everywhere are cringeworthy in terms of evaluation and reporting...
see also, this article titled bluntly "External validation of new risk prediction models is infrequent and reveals worse prognostic discrimination": https://www.sciencedirect.com/science/article/pii/S0895435614003539
