Latent Class Cluster Analysis: Selecting the number of clusters

Lezhnina, Olga; Kismihók, Gábor

doi:10.1016/j.mex.2022.101747

Cited by 20 publications

(9 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A known limitation of latent class analysis for cluster analysis is that it is computationally expensive, which might be inconvenient with large datasets. 48 Moreover, selecting the number of clusters is a challenging task involving inevitable subjective analytical choices. 48 With the principal component analysis method, however, we reproduced the two most commonly reported clusters of multimorbidity (mental health disorders and cardiometabolic disease), which shows the strength of the principal component analysis method.…”

Section: Discussionmentioning

confidence: 99%

Familial aggregation of multimorbidity in Sweden: national explorative family study

Zöller

Pirouzifard

Holmquist

et al. 2023

bmjmed

View full text Add to dashboard Cite

ObjectivesTo examine whether multimorbidity aggregates in families in Sweden.DesignNational explorative family study.SettingSwedish Multigeneration Register linked to the National Patient Register, 1997-2015. Multimorbidity was assessed with a modified counting method of 45 chronic non-communicable diseases according to ICD-10 (international classification of diseases, 10th revision) diagnoses.Participants2 694 442 Swedish born individuals (48.73% women) who could be linked to their Swedish born first, second, and third degree relatives. Twins were defined as full siblings born on the same date.Main outcome measuresMultimorbidity was defined as two or more non-communicable diseases. Familial associations for one, two, three, four, and five or more non-communicable diseases were assessed to examine risks depending on the number of non-communicable diseases. Familial adjusted odds ratios for multimorbidity were calculated for individuals with a diagnosis of multimorbidity compared with relatives of individuals unaffected by multimorbidity (reference). An initial principal component decomposition followed by a factor analysis with a principal factor method and an oblique promax rotation was used on the correlation matrix of tetrachoric correlations between 45 diagnoses in patients to identify disease clusters.ResultsThe odds ratios for multimorbidity were 2.89 in twins (95% confidence interval 2.56 to 3.25), 1.81 in full siblings (1.78 to 1.84), 1.26 in half siblings (1.24 to 1.28), and 1.13 in cousins (1.12 to 1.14) of relatives with a diagnosis of multimorbidity. The odds ratios for multimorbidity increased with the number of diseases in relatives. For example, among twins, the odds ratios for multimorbidity were 1.73, 2.84, 4.09, 4.63, and 6.66 for an increasing number of diseases in relatives, from one to five or more, respectively. Odds ratios were highest at younger ages: in twins, the odds ratio was 3.22 for those aged ≤20 years, 3.14 for those aged 21-30 years, and 2.29 for those aged >30 years at the end of follow-up. Nine disease clusters (factor clusters 1-9) were identified, of which seven aggregated in families. The first three disease clusters in the principal component decomposition were cardiometabolic disease (factor 1), mental health disorders (factor 2), and disorders of the digestive system (factor 3). Odds ratios for multimorbidity in twins, siblings, half siblings, and cousins for the factor 1 cluster were 2.79 (95% confidence interval 0.97 to 8.06), 2.62 (2.39 to 2.88), 1.52 (1.34 to 1.73), and 1.31 (1.23 to 1.39), and for the factor 2 cluster, 5.79 (4.48 to 7.48) 3.24 (3.13 to 3.36), 1.51 (1.45 to 1.57), and 1.37 (1.341.40).ConclusionsThe results of this explorative family study indicated that multimorbidity aggregated in Swedish families. The findings suggest that map clusters of diseases should be used for the genetic study of common diseases to show new genetic patterns of non-communicable diseases.

show abstract

Section: Discussionmentioning

confidence: 99%

Familial aggregation of multimorbidity in Sweden: national explorative family study

Zöller

Pirouzifard

Holmquist

et al. 2023

bmjmed

View full text Add to dashboard Cite

show abstract

“…Thus, each cluster has a probability of association, rather than a clear membership assignment. Due to the probabilistic nature of the class assignment, it may be difficult to derive instance-level associations, thus a single instance may belong marginally to multiple classes 32 , 33 . Alternatively, k-means allows for a characterization of clusters driven by centroid-based distances, allowing for a quantitative estimate of the membership 34 .…”

Section: Discussionmentioning

confidence: 99%

Cluster analysis driven by unsupervised latent feature learning of medications to identify novel pharmacophenotypes of critically ill patients

Sikora,

Jeong,

et al. 2023

Sci Rep

View full text Add to dashboard Cite

Unsupervised clustering of intensive care unit (ICU) medications may identify unique medication clusters (i.e., pharmacophenotypes) in critically ill adults. We performed an unsupervised analysis with Restricted Boltzmann Machine of 991 medications profiles of patients managed in the ICU to explore pharmacophenotypes that correlated with ICU complications (e.g., mechanical ventilation) and patient-centered outcomes (e.g., length of stay, mortality). Six unique pharmacophenotypes were observed, with unique medication profiles and clinically relevant differences in ICU complications and patient-centered outcomes. While pharmacophenotypes 2 and 4 had no statistically significant difference in ICU length of stay, duration of mechanical ventilation, or duration of vasopressor use, their mortality differed significantly (9.0% vs. 21.9%, p < 0.0001). Pharmacophenotype 4 had a mortality rate of 21.9%, compared with the rest of the pharmacophenotypes ranging from 2.5 to 9%. Phenotyping approaches have shown promise in classifying the heterogenous syndromes of critical illness to predict treatment response and guide clinical decision support systems but have never included comprehensive medication information. This first-ever machine learning approach revealed differences among empirically-derived subgroups of ICU patients that are not typically revealed by traditional classifiers. Identification of pharmacophenotypes may enable enhanced decision making to optimize treatment decisions.

show abstract

“…We subsequently evaluated the performance of each model, with the objective of determining the optimal fit for the data and the greatest possible distinction between the identified clusters. We utilized several statistical measures to evaluate the quality of the model fit, including the log likelihood plot, which indicates the point at which the log likelihood ceases to increase significantly, and the elbow heuristic for the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), where the change in successive values becomes less noticeable (Supplementary Table S3 and Figure S1) [18][19][20][21]. To gauge the extent of the distinction between latent clusters, an entropy value was used, where a value of ≥0.6 was indicative of favorable separation between the groups [21,22].…”

Section: Discussionmentioning

confidence: 99%

Infant Feeding Pattern Clusters Are Associated with Childhood Health Outcomes

Kim

Lee

et al. 2023

Nutrients

View full text Add to dashboard Cite

(1) Background: Feeding behavior habits have a pattern with a certain tendency during infancy. We aimed to identify the associations between feeding patterns in infancy and the subsequent 10-year childhood disease burden. (2) Methods: Data from 236,372 infants were obtained from the national health insurance and screening program records in South Korea. Parent-administered questionnaires during infancy provided details on the feeding type and types/frequency of complementary food for analyzing feeding patterns. The outcomes were all-cause hospitalization and the development of 15 representative childhood diseases until the age of 10 years. Anthropometric measurements obtained at 6 years of age were analyzed. To estimate outcome risks while considering multiple risk factors, we employed a Cox proportional hazard and modified Poisson regression. (3) Results: Three clusters were identified: high prevalence of breastfeeding with regular exposure to a variety of solid foods (n = 116,372, cluster 1), similar prevalence of breastfeeding and formula feeding with less exposure to solid foods (n = 108,189, cluster 2), and similar prevalence of breastfeeding and formula feeding with the least exposure to solid foods in infancy (n = 11,811, cluster 3). Compared with cluster 1, children in clusters 2 and 3 had increased risks of all-cause hospitalization (hazard ratio (HR), (95% confidence interval (CI)), 1.04 (1.03–1.06) and 1.08 (1.05–1.11), respectively). Children in clusters 2 and 3 had an increased risk of upper respiratory infection, pneumonia, and gastroenteritis, as well as neurobehavioral diseases. Overweight/obesity at the age of 6 years was associated with clusters 2 and 3. (4) Conclusions: Feeding patterns in infancy were associated with an increased risk of childhood disease burden.

show abstract

Latent Class Cluster Analysis: Selecting the number of clusters

Cited by 20 publications

References 18 publications

Familial aggregation of multimorbidity in Sweden: national explorative family study

Familial aggregation of multimorbidity in Sweden: national explorative family study

Cluster analysis driven by unsupervised latent feature learning of medications to identify novel pharmacophenotypes of critically ill patients

Infant Feeding Pattern Clusters Are Associated with Childhood Health Outcomes

Contact Info

Product

Resources

About