2022-04-18

Background

About heart failure

Heart disease is one of the leading causes of death with one in every four death due to cardiovascular disease in the U.S.

Heart failure is a subtype of chronic heart disease.

  • cause: insufficient blood and oxygen pumped into the circulation

  • implication: fatigue, shortness of breath, severe discomfort when carrying out physical activity, severe heart failure can directly cause death

  • prevalence: in 2018, heart failure was mentioned on 13.4% of death certificate

  • risk factors: hypertension, obesity, smoking, lacking physical activity, and excessive alcohol drinking

The objective of this analysis is to explore the relationship between survival following advanced heart failure and comorbidities.

About the data source

Data originally collected from 299 heart failure patients admitted to Institute of Cardiology and Allied hospital Faisalabad-Pakistan between April and December of 2015.

  • All patients had stage III or IV heart failure according to New York Heart Association classification.
 - Stage III means significant limitation in activity due to symptoms, and symtoms ease only at rest.

 - Stage IV means severe limitation in activity due to symptoms, and symtoms continue even at rest.

  • Patients were followed for an average of 130 days (range = 4-285 days) after being admitted to the hospital. The survival data were right-censored.

  • In total 96 patients died during the study (event) while the remaining 203 patients were censored at the end of the study.

  • The outcome is whether the patient survived till the end of the study, and the time variable is the time between hospital admission and either an event or a censoring point.

Data description

Patient characteristics

These three patient characteristics are the controlling covariates.

Variable Overall, N = 2991 censored, N = 2031 event, N = 961
age 60 (51, 70) 60 (50, 65) 65 (55, 75)
sex
women 105 (35%) 71 (35%) 34 (35%)
men 194 (65%) 132 (65%) 62 (65%)
smoking
non-smoker 203 (68%) 137 (67%) 66 (69%)
smoker 96 (32%) 66 (33%) 30 (31%)

1 Median (IQR); n (%)

Comorbidities

The comorbidities are the primary covariates of this analysis.

Variable Overall, N = 2991 censored, N = 2031 event, N = 961
anemia
non-anemic 170 (57%) 120 (59%) 50 (52%)
anemic 129 (43%) 83 (41%) 46 (48%)
diabetes
non-diabetic 174 (58%) 118 (58%) 56 (58%)
diabetic 125 (42%) 85 (42%) 40 (42%)
hypertension
non-hypertension 194 (65%) 137 (67%) 57 (59%)
hypertension 105 (35%) 66 (33%) 39 (41%)

1 n (%)

Data description continued

Other clinical measures collected at hospital admission

These clinical measures were treated as secondary covariates.

Variable Overall, N = 2991 censored, N = 2031 event, N = 961
creatine phosphokinase 250 (116, 582) 245 (109, 582) 259 (129, 582)
serum creatinine 1.10 (0.90, 1.40) 1.00 (0.90, 1.20) 1.30 (1.08, 1.90)
serum sodium 137.0 (134.0, 140.0) 137.0 (135.5, 140.0) 135.5 (133.0, 138.2)

1 Median (IQR)

  • Creatine phosphokinase (CPK): an indicator of heart/muscle stress or injury.

  • Serum creatinine: an indicator of kidney function

  • Serum sodium: an indicator of dehydration or kidney function.

Analysis

Plan

  1. Describe the overall survival using Kaplan-Meier (KM) method.

  2. Compare the survivorship by patient groups as defined by each of the comorbidities using stratified KM method.

  3. Use Cox proportional hazard (PH) model to estimate effect of each of the primary and secondary covariates, after adjusting for patient characteristics (controlling factors), on the survival following hospital admission.

  4. Construct a full Cox PH model containing all primary/secondary covariates that were found to be significant in step 3. Run a model selection process to make an efficient final model.

  5. All Cox PH models were examined for assumptions of proportionality and linearity.

Overall survival

The overall survival did not show a sudden drop, and it did not reach 50 percentile (median survival).

Stratified survival - by comorbidities

anemia

The two survival curves seem to separate, and anemic patients showed a consistently-lower survivorship than non-anemic patients. However, the log-rank test was not significant at 0.05.

Stratified survival - by comorbidities

diabetes

The two survival curves intertwined with each other, indicating similar survival between those diabetic and non-diabetic patients. This is also confirmed by the log-rank test p-value.

Stratified survival - by comorbidities

hypertension

The two curves separated well, and the log-rank p-value also indicates a significantly better survival among those patients do not have hypertension than those who between patients who have hypertension.

Adjusted Cox PH model - effect of anemia

Cox PH model output

## # A tibble: 5 x 3
##   coef                  HR_CI            correlation_test_p
##   <chr>                 <chr>                         <dbl>
## 1 age                   1.04 (1.03-1.06)              0.378
## 2 factor(sex)men        0.97 (0.61-1.55)              0.531
## 3 factor(smoking)smoker 1.09 (0.67-1.77)              0.473
## 4 factor(anemia)anemic  1.33 (0.89-2.00)              0.499
## 5 global                <NA>                          0.739

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - effect of diabetes

Cox PH model output

## # A tibble: 5 x 3
##   coef                     HR_CI            correlation_test_p
##   <chr>                    <chr>                         <dbl>
## 1 age                      1.04 (1.03-1.06)              0.381
## 2 factor(sex)men           0.97 (0.61-1.55)              0.573
## 3 factor(smoking)smoker    1.08 (0.66-1.75)              0.522
## 4 factor(diabetes)diabetic 1.13 (0.74-1.72)              0.918
## 5 global                   <NA>                          0.821

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - effect of hypertension

Cox PH model output

## # A tibble: 5 x 3
##   coef                             HR_CI            correlation_test_p
##   <chr>                            <chr>                         <dbl>
## 1 age                              1.04 (1.03-1.06)              0.394
## 2 factor(sex)men                   0.99 (0.62-1.58)              0.534
## 3 factor(smoking)smoker            1.09 (0.67-1.77)              0.522
## 4 factor(hypertension)hypertension 1.53 (1.01-2.31)              0.561
## 5 global                           <NA>                          0.768

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - effect of creatine phosphokinase (CPK)

Cox PH model output

## # A tibble: 5 x 3
##   coef                          HR_CI            correlation_test_p
##   <chr>                         <chr>                         <dbl>
## 1 age                           1.04 (1.03-1.06)              0.374
## 2 factor(sex)men                0.96 (0.60-1.54)              0.577
## 3 factor(smoking)smoker         1.06 (0.65-1.73)              0.523
## 4 log(`creatine phosphokinase`) 1.01 (0.84-1.21)              0.129
## 5 global                        <NA>                          0.479

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - effect of serum creatinine

Cox PH model output

## # A tibble: 5 x 3
##   coef                    HR_CI            correlation_test_p
##   <chr>                   <chr>                         <dbl>
## 1 age                     1.04 (1.02-1.06)              0.535
## 2 factor(sex)men          0.93 (0.58-1.49)              0.519
## 3 factor(smoking)smoker   1.18 (0.72-1.92)              0.352
## 4 log(`serum creatinine`) 2.74 (1.94-3.88)              0.311
## 5 global                  <NA>                          0.651

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - effect of serum sodium

Cox PH model output

## # A tibble: 5 x 3
##   coef                  HR_CI            correlation_test_p
##   <chr>                 <chr>                         <dbl>
## 1 age                   1.04 (1.03-1.06)              0.505
## 2 factor(sex)men        0.93 (0.58-1.49)              0.631
## 3 factor(smoking)smoker 1.06 (0.65-1.73)              0.487
## 4 `serum sodium`        0.93 (0.90-0.97)              0.615
## 5 global                <NA>                          0.859

Plot of the correlation test

Cumulative hazard plot

Adjusted Cox PH model - summary

Based on results from the previously observed Cox PH hazard ratios, four candidate factors made to the final competition, namely: hypertension, serum creatine, serum sodium, and the patient characteristics.

AIC-based model selection

The AIC-based model selection process automatically dropped sex and smoking

  • full model: survival = age + sex + smoking + hypertension + log(serum creatine) + serum sodium

  • final model: survival = age + hypertension + log(serum creatine) + serum sodium

Final model output

## # A tibble: 5 x 3
##   coef                             HR_CI            PH_test_pvalue
##   <chr>                            <chr>                     <dbl>
## 1 age                              1.04 (1.02-1.06)          0.709
## 2 `serum sodium`                   0.95 (0.91-0.99)          0.631
## 3 log(`serum creatinine`)          2.46 (1.73-3.51)          0.357
## 4 factor(hypertension)hypertension 1.66 (1.10-2.51)          0.576
## 5 global                           <NA>                      0.717

Conclusion: Hypertension and kidney functions were associated with survival following advanced heart failure.

Evaluating PH assumptions for the final model - correlation test plot

Evaluating PH assumptions for the final model - Schoenfeld residuals

Evaluating PH assumptions for the final model - Martingale residuals

Appendix

Dataset:

Reference

Data source: https://www.kaggle.com/andrewmvd/heart-failure-clinical-data

Original study that published on this data: Survival analysis of heart failure patients: A case study

Other published article on this data: Machine learning can predict survival of patients with heart failure from serum creatinine and ejection fraction alone

Virani SS, Alonso A, Aparicio HJ, et al. Heart Disease and Stroke Statistics-2021 Update: A Report From the American Heart Association. Circulation. 2021;143(8):e254-e743. doi:10.1161/CIR.0000000000000950

Centers for Disease Control and Prevention. Heart Failure. September 8, 2020. Accessed March 29th, 2022. https://www.cdc.gov/heartdisease/heart_failure.htm

Shah R, Agarwal AK. Anemia associated with chronic heart failure: current concepts. Clin Interv Aging. 2013;8:111-122. doi:10.2147/CIA.S27105

Reference continued

Zareini B, Blanche P, D’Souza M, et al. Type 2 Diabetes Mellitus and Impact of Heart Failure on Prognosis Compared to Other Cardiovascular Diseases: A Nationwide Study. Circ Cardiovasc Qual Outcomes. 2020;13(7):e006260. doi:10.1161/CIRCOUTCOMES.119.006260

Tackling G, Borhade MB. Hypertensive Heart Disease. In: StatPearls. Treasure Island (FL): StatPearls Publishing; July 1, 2021.

Ahmad T, Munir A, Bhatti SH, Aftab M, Raza MA. Survival analysis of heart failure patients: A case study. PLoS One. 2017;12(7):e0181001. Published 2017 Jul 20. doi:10.1371/journal.pone.0181001

Heart. Classes of Heart Failure. May 31, 2017. Accessed March 27th, 2022. https://www.heart.org/en/health-topics/heart-failure/what-is-heart-failure/classes-of-heart-failure

Widmer F. Herzinsuffizienz und Komorbiditäten [Comorbidity in heart failure]. Ther Umsch. 2011;68(2):103-106. doi:10.1024/0040-5930/a000127