A novel prediction model for curve progression to surgical threshold in adolescent idiopathic scoliosis derived from unsupervised machine learning of bone microarchitecture phenotypes – a 6-year longitudinal study of 323 patients followed till skeletal maturity
Invited conference paper presented and published in conference proceedings
CUHK Authors
Full Text
There are no full text file(s) associated with this record. |
Other information
AbstractIntroduction:
Progressive AIS can lead to serious complications and morbidity. Models to predict therisk of curve progression in AIS were developed with suboptimal results. Poor bone qualities are asignificant prognostic factor for curve progression in AIS. This study aims to(a) investigate theassociation between unsupervised-machine learning-identified subtypes of bone microarchitecture phenotypes and risk of curve progression, and(b) to assess the accuracy of bone-microarchitecture-phenotype-based prediction models for curve progression to surgical threshold in AIS longitudinally tillskeletal maturity.
Materials and Methods:
Female patients(11-14y.o.) with AIS were recruited and theirbone qualities were evaluated with high-resolution peripheral quantitative computed tomography(HR-pQCT). Three bone microarchitecture phenotypes were identified by fuzzy c-means with their risk ofcurve progression to a Cobb angle≥50º being assessed with Cox’s proportional hazards models.Accuracy of logistic regression models in prediction of curve progression to a Cobb angle≥50º tillskeletal maturity(16y.o.) in AIS based on the identified phenotypes with(Model-1) and without(Model-2)baseline Cobb angle was evaluated with area under the curve(AUC).
Results:
A total of 323 female AISpatients(12.86±0.88y.o.) with a Cobb angle of 26.08±8.86º were recruited. When compared withPhenotype-1, hazard ratio in Phenotype-2 and Phenotype-3 were1.32(p=0.608) and 3.05(p=0.024),respectively. AUC of Model-1 and Model-2 were 0.937 and 0.863, respectively.
Discussion andconclusion:
Bone microarchitecture phenotypes derived from unsupervised machine learning aresignificantly associated with risk of curve progression in AIS. The identified bone microarchitecturephenotypes are useful in developing highly accurate prediction models for curve progression tosurgical threshold in AIS.
Progressive AIS can lead to serious complications and morbidity. Models to predict therisk of curve progression in AIS were developed with suboptimal results. Poor bone qualities are asignificant prognostic factor for curve progression in AIS. This study aims to(a) investigate theassociation between unsupervised-machine learning-identified subtypes of bone microarchitecture phenotypes and risk of curve progression, and(b) to assess the accuracy of bone-microarchitecture-phenotype-based prediction models for curve progression to surgical threshold in AIS longitudinally tillskeletal maturity.
Materials and Methods:
Female patients(11-14y.o.) with AIS were recruited and theirbone qualities were evaluated with high-resolution peripheral quantitative computed tomography(HR-pQCT). Three bone microarchitecture phenotypes were identified by fuzzy c-means with their risk ofcurve progression to a Cobb angle≥50º being assessed with Cox’s proportional hazards models.Accuracy of logistic regression models in prediction of curve progression to a Cobb angle≥50º tillskeletal maturity(16y.o.) in AIS based on the identified phenotypes with(Model-1) and without(Model-2)baseline Cobb angle was evaluated with area under the curve(AUC).
Results:
A total of 323 female AISpatients(12.86±0.88y.o.) with a Cobb angle of 26.08±8.86º were recruited. When compared withPhenotype-1, hazard ratio in Phenotype-2 and Phenotype-3 were1.32(p=0.608) and 3.05(p=0.024),respectively. AUC of Model-1 and Model-2 were 0.937 and 0.863, respectively.
Discussion andconclusion:
Bone microarchitecture phenotypes derived from unsupervised machine learning aresignificantly associated with risk of curve progression in AIS. The identified bone microarchitecturephenotypes are useful in developing highly accurate prediction models for curve progression tosurgical threshold in AIS.
Acceptance Date15/09/2022
All Author(s) ListKenneth Yang, Wayne Lee, Alec Hung, Kumar Anubrat, Raymond Wan, Jack Cheng, Tsz-Ping Lam
Name of ConferenceThe Hong Kong Orthopaedic Association 2022 Annual Congress
Start Date of Conference05/11/2022
End Date of Conference06/11/2022
Place of ConferenceHong Kong
Country/Region of ConferenceHong Kong
Year2022
LanguagesEnglish-United States
