In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm

Linda Nichols; Tom Taverner; Francesca Crowe; Sylvia Richardson; Christopher Yau; Steven Kiddle; Paul Kirk; Jessica Barrett; Krishnarajah Nirantharakumar; Simon Griffin; Duncan Edwards; Tom Marshall

doi:10.1016/j.jclinepi.2022.10.011

In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm

Linda Nichols, Tom Taverner, Francesca Crowe, Sylvia Richardson, Christopher Yau, Steven Kiddle, Paul Kirk, Jessica Barrett, Krishnarajah Nirantharakumar, Simon Griffin, Duncan Edwards, Tom Marshall

Applied Health Research

Research output: Contribution to journal › Article › peer-review

28 Downloads (Pure)

Abstract

BACKGROUND AND OBJECTIVES: To investigate the reproducibility and validity of latent class analysis (LCA) and hierarchical cluster analysis (HCA), multiple correspondence analysis followed by k-means (MCA-kmeans) and k-means (kmeans) for multimorbidity clustering.

METHODS: We first investigated clustering algorithms in simulated datasets with 26 diseases of varying prevalence in predetermined clusters, comparing the derived clusters to known clusters using the adjusted Rand Index (aRI). We then them investigated in the medical records of male patients, aged 65 to 84 years from 50 UK general practices, with 49 long-term health conditions. We compared within cluster morbidity profiles using the Pearson correlation coefficient and assessed cluster stability was in 400 bootstrap samples.

RESULTS: In the simulated datasets, the closest agreement (largest aRI) to known clusters was with LCA and then MCA-kmeans algorithms. In the medical records dataset, all four algorithms identified one cluster of 20-25% of the dataset with about 82% of the same patients across all four algorithms. LCA and MCA-kmeans both found a second cluster of 7% of the dataset. Other clusters were found by only one algorithm. LCA and MCA-kmeans clustering gave the most similar partitioning (aRI 0.54).

CONCLUSION: LCA achieved higher aRI than other clustering algorithms.

Original language	English
Pages (from-to)	164-175
Number of pages	12
Journal	Journal of Clinical Epidemiology
Volume	152
Early online date	11 Oct 2022
DOIs	https://doi.org/10.1016/j.jclinepi.2022.10.011
Publication status	E-pub ahead of print - 11 Oct 2022

Bibliographical note

Keywords

Multimorbidity
Clustering methods
Electronic medical records
latent class analysis
Hierarchical cluster analysis
HierarchMultiple correspondence analysi
K-means

Access to Document

10.1016/j.jclinepi.2022.10.011Licence: Creative Commons: Attribution (CC BY)

Nichols2022_In_simulated_data_and_health_recordsFinal published version, 950 KBLicence: Creative Commons: Attribution (CC BY)

Bringing Innovative Research Methods to Clustering Analysis of Multimorbidity (BIRM-CAM)
Crowe, F., Marshall, T., Yau, C. & Nirantharakumar, K.
Medical Research Council
1/10/19 → 30/06/23
Project: Research Councils

Cite this

Nichols, L., Taverner, T., Crowe, F., Richardson, S., Yau, C., Kiddle, S., Kirk, P., Barrett, J., Nirantharakumar, K., Griffin, S., Edwards, D., & Marshall, T. (2022). In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm. Journal of Clinical Epidemiology, 152, 164-175. Advance online publication. https://doi.org/10.1016/j.jclinepi.2022.10.011

@article{1355824018334880b3b9fa354b245fba,

title = "In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm",

abstract = "BACKGROUND AND OBJECTIVES: To investigate the reproducibility and validity of latent class analysis (LCA) and hierarchical cluster analysis (HCA), multiple correspondence analysis followed by k-means (MCA-kmeans) and k-means (kmeans) for multimorbidity clustering.METHODS: We first investigated clustering algorithms in simulated datasets with 26 diseases of varying prevalence in predetermined clusters, comparing the derived clusters to known clusters using the adjusted Rand Index (aRI). We then them investigated in the medical records of male patients, aged 65 to 84 years from 50 UK general practices, with 49 long-term health conditions. We compared within cluster morbidity profiles using the Pearson correlation coefficient and assessed cluster stability was in 400 bootstrap samples.RESULTS: In the simulated datasets, the closest agreement (largest aRI) to known clusters was with LCA and then MCA-kmeans algorithms. In the medical records dataset, all four algorithms identified one cluster of 20-25% of the dataset with about 82% of the same patients across all four algorithms. LCA and MCA-kmeans both found a second cluster of 7% of the dataset. Other clusters were found by only one algorithm. LCA and MCA-kmeans clustering gave the most similar partitioning (aRI 0.54).CONCLUSION: LCA achieved higher aRI than other clustering algorithms.",

keywords = "Multimorbidity, Clustering methods, Electronic medical records, latent class analysis, Hierarchical cluster analysis, HierarchMultiple correspondence analysi, K-means",

author = "Linda Nichols and Tom Taverner and Francesca Crowe and Sylvia Richardson and Christopher Yau and Steven Kiddle and Paul Kirk and Jessica Barrett and Krishnarajah Nirantharakumar and Simon Griffin and Duncan Edwards and Tom Marshall",

year = "2022",

month = oct,

day = "11",

doi = "10.1016/j.jclinepi.2022.10.011",

language = "English",

volume = "152",

pages = "164--175",

journal = "Journal of Clinical Epidemiology",

issn = "0895-4356",

publisher = "Elsevier",

}

Nichols, L, Taverner, T, Crowe, F, Richardson, S, Yau, C, Kiddle, S, Kirk, P, Barrett, J, Nirantharakumar, K, Griffin, S, Edwards, D & Marshall, T 2022, 'In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm', Journal of Clinical Epidemiology, vol. 152, pp. 164-175. https://doi.org/10.1016/j.jclinepi.2022.10.011

TY - JOUR

T1 - In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm

AU - Nichols, Linda

AU - Taverner, Tom

AU - Crowe, Francesca

AU - Richardson, Sylvia

AU - Yau, Christopher

AU - Kiddle, Steven

AU - Kirk, Paul

AU - Barrett, Jessica

AU - Nirantharakumar, Krishnarajah

AU - Griffin, Simon

AU - Edwards, Duncan

AU - Marshall, Tom

PY - 2022/10/11

Y1 - 2022/10/11

N2 - BACKGROUND AND OBJECTIVES: To investigate the reproducibility and validity of latent class analysis (LCA) and hierarchical cluster analysis (HCA), multiple correspondence analysis followed by k-means (MCA-kmeans) and k-means (kmeans) for multimorbidity clustering.METHODS: We first investigated clustering algorithms in simulated datasets with 26 diseases of varying prevalence in predetermined clusters, comparing the derived clusters to known clusters using the adjusted Rand Index (aRI). We then them investigated in the medical records of male patients, aged 65 to 84 years from 50 UK general practices, with 49 long-term health conditions. We compared within cluster morbidity profiles using the Pearson correlation coefficient and assessed cluster stability was in 400 bootstrap samples.RESULTS: In the simulated datasets, the closest agreement (largest aRI) to known clusters was with LCA and then MCA-kmeans algorithms. In the medical records dataset, all four algorithms identified one cluster of 20-25% of the dataset with about 82% of the same patients across all four algorithms. LCA and MCA-kmeans both found a second cluster of 7% of the dataset. Other clusters were found by only one algorithm. LCA and MCA-kmeans clustering gave the most similar partitioning (aRI 0.54).CONCLUSION: LCA achieved higher aRI than other clustering algorithms.

AB - BACKGROUND AND OBJECTIVES: To investigate the reproducibility and validity of latent class analysis (LCA) and hierarchical cluster analysis (HCA), multiple correspondence analysis followed by k-means (MCA-kmeans) and k-means (kmeans) for multimorbidity clustering.METHODS: We first investigated clustering algorithms in simulated datasets with 26 diseases of varying prevalence in predetermined clusters, comparing the derived clusters to known clusters using the adjusted Rand Index (aRI). We then them investigated in the medical records of male patients, aged 65 to 84 years from 50 UK general practices, with 49 long-term health conditions. We compared within cluster morbidity profiles using the Pearson correlation coefficient and assessed cluster stability was in 400 bootstrap samples.RESULTS: In the simulated datasets, the closest agreement (largest aRI) to known clusters was with LCA and then MCA-kmeans algorithms. In the medical records dataset, all four algorithms identified one cluster of 20-25% of the dataset with about 82% of the same patients across all four algorithms. LCA and MCA-kmeans both found a second cluster of 7% of the dataset. Other clusters were found by only one algorithm. LCA and MCA-kmeans clustering gave the most similar partitioning (aRI 0.54).CONCLUSION: LCA achieved higher aRI than other clustering algorithms.

KW - Multimorbidity

KW - Clustering methods

KW - Electronic medical records

KW - latent class analysis

KW - Hierarchical cluster analysis

KW - HierarchMultiple correspondence analysi

KW - K-means

U2 - 10.1016/j.jclinepi.2022.10.011

DO - 10.1016/j.jclinepi.2022.10.011

M3 - Article

C2 - 36228971

SN - 0895-4356

VL - 152

SP - 164

EP - 175

JO - Journal of Clinical Epidemiology

JF - Journal of Clinical Epidemiology

ER -

In simulated data and health records, latent class analysis was the optimum multimorbidity clustering algorithm

Abstract

Bibliographical note

Keywords

Access to Document

Fingerprint

Projects

Bringing Innovative Research Methods to Clustering Analysis of Multimorbidity (BIRM-CAM)

Cite this