PPubMed5 Feb 2016Cited 103×

Fusing Data Mining, Machine Learning and Traditional Statistics to Detect Biomarkers Associated with Depression Data Mine & Machine Learn Biomarkers of Depression

Joanna F. Dipnall, Julie A. Pasco, Michael Berk, Lana J. Williams, Seetal Dodd, Felice N. Jacka, Denny Meyer, Mansour Ebrahimi

Abstract

Running a multiple regression directly on the 21 variables selected from the boosted regression on the training and validation data sets, without first using univariate regressions to reduce the number of variables further, produced some unstable results across the training and validation data sets. The odds ratio for Haemoglobin differed drastically between the two models which was consistent with haemoglobin having singularity with haematocrit (r = 0.968) . In addition, predictor significance was inconsistent: total bilirubin was the only significant predictor in the training model and no pr

A figure from Fusing Data Mining, Machine Learning and Traditional Statistics to Detect Biomarkers Associated with Depression Data Mine & Machine Learn Biomarkers of Depression — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

A hybrid data‑mining and statistical method identified red cell distribution width, serum glucose, and total bilirubin as depression biomarkers in NHANES.

What’s new

i.Developed a three‑step hybrid methodology that merges data mining, machine learning (boosted regression), and traditional survey‑weighted logistic regression.
ii.Implemented multiple imputation to handle missing data across 67 biomarkers in a complex NHANES survey design.
iii.Identified three biomarkers—red cell distribution width, serum glucose, and total bilirubin—as significant predictors of depression.

How they did it

The study first applied multiple imputation (20 chained data sets) to address missing values, then used a boosted regression tree algorithm with bagging and shrinkage to rank variable importance, and finally fitted a survey‑weighted multivariate logistic regression on the top predictors. The approach accounted for NHANES's complex four‑stage sampling design throughout.

What they found

Final analytic sample: 5,227 participants (combined original and imputed data sets: 109,767 observations).
Twenty‑one biomarkers were selected after boosted regression, explaining >50% of total relative importance (53.85% original, 53.33% across imputed sets).
Three biomarkers remained significant in the final model (p<0.05): red cell distribution width, serum glucose, and total bilirubin.
Goodness‑of‑fit test for the final model was non‑significant (F(9,8)=2.05, p=0.163), indicating an adequate fit.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Similar papers in all four archives

The same idea searched in arXiv, PubMed, bioRxiv and chemRxiv at once, so a method can lead you to where it is being applied.

AarXiv4 found

PPubMed3 found

BbioRxiv4 found

CchemRxiv2 found

Searching the other archives…