Unveiling unique clinical phenotypes of hip fracture patients and the temporal association with cardiovascular events

Cardiovascular events are the leading cause of death among hip fracture patients. This study aims to identify subphenotypes of hip fracture patients and investigate their association with incident cardiovascular events, all-cause mortality, and health service utilisation in Hong Kong and the United Kingdom populations. By the latent class analysis, we show three distinct clusters in the Hong Kong cohort (n = 78,417): Cluster 1 has cerebrovascular and hypertensive diseases, hyperlipidemia, and diabetes; Cluster 2 has congestive heart failure; Cluster 3 consists of relatively healthy patients. Compared to Cluster 3, higher risks of major adverse cardiovascular events are observed in Cluster 1 (hazard ratio 1.97, 95% CI 1.83 to 2.12) and Cluster 2 (hazard ratio 4.06, 95% CI 3.78 to 4.35). Clusters 1 and 2 are also associated with a higher risk of mortality, more unplanned accident and emergency visits and longer hospital stays. Self-controlled case series analysis shows a significantly elevated risk of major adverse cardiovascular events within 60 days post-hip fracture. Similar associations are observed in the United Kingdom cohort (n = 27,948). Pre-existing heart failure is identified as a unique subphenotype associated with poor prognosis after hip fractures.


Supplementary Table 2. Summary of class membership probability.
Class membership probabilities are computed for each subject through LCA.This probability represents the likelihood that a subject would be assigned to a certain class (computed using the subject's profile defined by the clustering variables).For example, within the context of a three-cluster LCA solution, each patient is associated with three class membership probabilities, corresponding to the likelihood of their belonging to each of the three latent classes.These probabilities quantify the confidence in class assignment, with higher values indicating greater certainty that a patient aligns with a particular class.Patients are assigned into the class for which they hold the highest membership probability.For each cluster identified, latent class analysis computed a set of conditional probabilities associated with the 22 clustering variables considered in this study.These conditional probabilities, also known as item-response probabilities, were estimated through the maximum likelihood estimation process in LCA.The item-response probabilities are the likelihood of observing a specific response given a latent class.For example, a conditional probability of 0.598 for baseline heart failure in cluster 2 in HK CDARS implies that there is a 59.8% chance that a patient classified to cluster 2 has baseline heart failure.
Each cluster could therefore be characterised by a list of 22 conditional probability values (one for each clustering variable), that could be leveraged for cluster comparison.For example, to assess the similarity between Cluster 1 in the training set and Cluster 1 in the test set, the Pearson's correlation coefficient was computed using their respective lists of 22 conditional probabilities.This process was repeated for each pair of clusters between the training and test sets.
Higher correlation coefficients towards 1 indicate higher degree of similarity between the two clusters being compared (a similar profile regarding the presence of the clustering variables).
The results from Supplementary BIC plot for the UK THIN cohort identified a marked inflection zone between 2 to 4 clusters.The absolute ASW value was the highest at 2 clusters, while the ICL value peaked at a single cluster, followed by a steady decline in ICL values with additional clusters.The downward trends in both ASW and ICL beyond two clusters suggested that a smaller number of clusters may better capture the data structure.Integrating all the three metrics led to our conclusion of evaluating a 2-clusters solution for the UK THIN cohort.(b) UK THIN Notes: MI = myocardial infarction; IRR = Incidence Rate Ratios; CI = Confidence Interval.(b) UK THIN

Supplementary Table 5. The association between hip fracture subphenotypes and 180-day outcomes of interest, excluding patients with MACE within 30 days after index date.
n = 77,056 independent patients in the HK CDARS after removal of patients with MACE within 30 days after index date.b n = 27,754 independent patients in the UK THIN after removal of patients with MACE within 30 days after index date.c Hazard Ratios, associated 95% CIs and two-sided p-values are derived from Cox proportional regression, with adjustment of age and sex.d Hazard Ratios, associated 95% CIs and two-sided p-values are derived from competing risk regression, with adjustment of age and sex.e Incidence Rate Ratios, associated 95% CIs and two-sided p-values are derived from Poisson regression, with adjustment of age and sex.
Notes: MACE = Major Adverse Cardiovascular Events; A&E = Accident and Emergency; CI = Confidence Interval.a

Table 6 . The association between hip fracture subphenotypes and 180-day cardiac A&E hospitalisation outcomes in HK CDARS.
Notes: MACE = Major Adverse Cardiovascular Events; CI = Confidence Interval.aHazardRatios, associated 95% CIs and two-sided p-values are derived from competing risk regression, with adjustment of age and sex.Supplementary Table 7.The association between hip fracture subphenotypes and 180-day MACE defined by myocardial infarction and stroke.Notes: MACE = Major Adverse Cardiovascular Events; MI = Myocardial Infarction; CI = Confidence Interval.a Hazard Ratios, associated 95% CIs and two-sided p-values are derived from competing risk regression, with adjustment of age and sex.

Table 8 . Stratified analysis on the association between hip fracture subphenotypes and 180-day outcomes of interest.
HR = Hazard Ratio; MACE = Major Adverse Cardiovascular Events; CI = Confidence Interval.aHazard Ratios, associated 95% CIs and two-sided p-values are derived from Cox proportional regression.bHazard Ratios, associated 95% CIs and two-sided p-values are derived from competing risk regression.Older age group defined by 85 years or above, whereas younger age group was defined by below 85 years.Analysis adjusted for sex.e Internal fixation (n = 24,526 independent patients) and partial hip replacement (n = 14,042 independent patients) were the two most common procedures recorded 7 days before or after the hip fracture event date.Analysis adjusted for age and sex.
Notes: c Analysis adjusted for age.d

Table 9 . Results of competing risk regression analysis on the association between hip fracture subphenotypes and risk of individual MACE outcomes at multiple time points. (a) HK CDARS
Data are presented as Hazard Ratio (95% Confidence Interval) two-sided p-value derived from competing risk regression.All analyses are adjusted for age and sex.
Note: Data are presented as Hazard Ratio (95% Confidence Interval) two-sided p-value derived from competing risk regression.All analyses are adjusted for age and sex.(b)UKTHINNote:

Table 10. Results of the SCCS analysis (individual MACE outcomes and sensitivity analysis). (a) HK CDARS
MACE = Major Adverse Cardiovascular Events; d = day; IRR = Incidence Rate Ratios; CI = Confidence Interval. Notes:

Considering only post-hip fracture period as the baseline: MACE
Baseline period indicates 366 days before hip fracture plus 181 to 732 days after hip fracture.c Baseline period indicates 181 to 732 days after hip fracture.
Notes: MACE = Major Adverse Cardiovascular Events; d = day; IRR = Incidence Rate Ratios; CI = Confidence Interval.a IRR, incidence rate ratio adjusted for age by quintiles.b 14 Supplementary

Table 11. Comparison of the event rates of the hip fracture subphenotypes with the MI reference cohort in HK CDARS.
MACE defined by MI and stroke (as heart failure hospitalisation data were not available in the MI Reference Cohort).
Notes: MACE = Major Adverse Cardiovascular; MI = Myocardial infarction.* The Incidence Rates are indicated by the symbols (circle for Cluster 1; triangle for Cluster 2).n = 27,948 independent patients in UK THIN (n = 4,966 in Cluster 1; n = 22,982 in Cluster 2).Note: MACE = Major Adverse Cardiovascular Events.
Incidence Rates for MACE (per 1000 Person-months)