📜Background

Heart failure remains a leading global killer, with South Asia facing unique challenges in left ventricular systolic dysfunction management. Punjab has a high-risk population with prevalent modifiable risk factors. Limited local data exists on combined clinical/biochemical survival predictors.

🎯Objectives & Aims

The aim of the study was to estimate death rates due to heart failure for patients who were admitted to Institute of Cardiology and Allied hospital Faisalabad-Pakistan during April-December (2015) and to Develop an integrated risk model with clinical/lab variables including ejection fraction, serum markers, and comorbidities.

🧪Study Design
Prospective cohort of 299 consecutive New York Heart Association functional class III or IV patients (LVEF<45%, age≥40). Combined echocardiography, clinical records, and lab data (CPK, creatinine, sodium) with minimum 6-month follow-up.
📊Justification
• Powered (80%) to detect HR≥1.5 (n=299)
• Represents regional demographics (35%♀/65%♂)
• First to combine clinical + biochemical predictors
• Addresses critical gap in South Asian HF data
📈Statistical Analysis
Descriptive: Mean±SD / Proportions / KM curves
Inferential: Chi-squared and t-tests for preliminary analysis
Cox Model: Stepwise selection (sequential replacement) for optimal predictors of mortality. Proportional hazards assumptions assessed using Schoenfeld residuals
Validation: AIC/BIC • LRT • Residuals analysis
Software: R 4.4.3 (survival, survminer and tidyverse packages)
Full Analysis Report

Definition of terms

  • Class III (Moderate): Marked limitation of physical activity. Comfortable at rest, but less than ordinary activity causes fatigue, palpitation, or dyspnea.
  • Class IV (Severe): Unable to carry out any physical activity without discomfort. Symptoms of cardiac insufficiency at rest.

Variable discription

⏱️TIME
Days from treatment to clinical event (death/transplant)
💓Event
Clinical outcomes: Alive (0) Death (1)
🚻Gender
Biological sex: Male (0), Female (1)
🚬Smoking
Current smoking status: Non-smoker (0), Smoker (1)
💉Diabetes
Diabetes mellitus diagnosis: Absent (0), Present (1)
🩸BP
Systolic blood pressure measured in mmHg
👤Age
Patient age at baseline in complete years
🩹Anaemia
Hemoglobin status: Normal (0, Hb ≥12g/dL women/≥13g/dL men), Anaemic (1)
❤️Ejection Fraction
Left ventricular ejection fraction percentage (normal range 55-70%)
🧪Sodium
Serum sodium level in mEq/L (normal range 135-145)
🧫Creatinine
Serum creatinine in mg/dL indicating kidney function
🔬Platelets
Platelet count in cells/μL (normal range 150,000-450,000)
🧪CPK
Creatinine phosphokinase enzyme level in U/L indicating heart muscle status
📝Note
All laboratory measurements represent baseline values prior to treatment initiation

Explanatory data analysis

Correlation Analysis

  • There were no significant correlations among the dependent variables.
  • hence no issues with multicollinearity would be detected in model building steps.

Frequencies and proportions

  • the majority of patients where in the 30 <EF<=45 category.
  • Non-smokers outnumbered smokers
  • More patients were non-diabetic
  • Majority had normal blood pressure
  • Non-anaemic patients predominated

Histograms

  • It can be noted that there is some significant differences in Ejection Fraction between between people who died and those who survived. From the boxplots ,the mean age seems to be higher for patients who died as compared to those who censored.
  • Other boxplots do not show any significant mean differences.

Hypothesis Testing

Characteristic alive
N = 2031		 dead
N = 961		 Test Statistic2 p-value2
BP

1.9 0.17
    No 137 / 203 (67%) 57 / 96 (59%)

    Yes 66 / 203 (33%) 39 / 96 (41%)

platelets_cat

3.5 0.17
    < Q1 61 / 203 (30%) 39 / 96 (41%)

    < Q2 73 / 203 (36%) 27 / 96 (28%)

    < Q3 69 / 203 (34%) 30 / 96 (31%)

EFraction_cat

32 <0.001
    30-45 115 / 203 (57%) 31 / 96 (32%)

    EF<=30 42 / 203 (21%) 51 / 96 (53%)

    EF>45 46 / 203 (23%) 14 / 96 (15%)

1 n / N (%)
2 Pearson’s Chi-squared test
Characteristic alive
N = 2031		 dead
N = 961		 Test Statistic2 p-value2
Diabetes

0.00 0.97
    No 118 / 203 (58%) 56 / 96 (58%)

    Yes 85 / 203 (42%) 40 / 96 (42%)

Anaemia

1.3 0.25
    No 120 / 203 (59%) 50 / 96 (52%)

    Yes 83 / 203 (41%) 46 / 96 (48%)

Gender

0.01 0.94
    F 71 / 203 (35%) 34 / 96 (35%)

    M 132 / 203 (65%) 62 / 96 (65%)

Smoking

0.05 0.83
    No 137 / 203 (67%) 66 / 96 (69%)

    Yes 66 / 203 (33%) 30 / 96 (31%)

1 n / N (%)
2 Pearson’s Chi-squared test

conclusions

Characteristic alive
N = 2031		 dead
N = 961		 Test Statistic2 p-value2
Age 59 (11) 65 (13) -4.2 <0.001
Sodium 137.2 (4.0) 135.4 (5.0) 3.2 0.002
Creatinine 1.18 (0.65) 1.84 (1.47) -4.2 <0.001
Ejection.Fraction 40 (11) 33 (13) 4.6 <0.001
Pletelets 266,657 (97,531) 256,381 (98,526) 0.84 0.40
CPK 540 (754) 670 (1,317) -0.90 0.37
1 Mean (SD)
2 Welch Two Sample t-test
  • Ejection fraction is significantly associated to mortality at 5% level of significance(\(\chi^2=32,p<0.001\).
  • There is no significant evidence to show association for other categorical predictors with mortality. - There is significant difference in mean Age of patients who died and those who survived (\(p<0.001\)).
  • There is statistically significant difference in mean Sodium and Creatinine for patients who died and those who survived (\(p=0.002\)) and (\(p<0.001\)) respectively.

Kaplan Meier Curves

  • The plots show the overall survival rates for the patients and here we notice that first event occured at day 4, with survival probability dropping to 99.7% ,between days 10-30 (survival drops from 96% to 88%) ,Median survival was not reached (since survival doesn’t drop below 50%).
  • The steepness continues until day 241 with final survival at 57.6%

Kaplan Meier Curves

  • The survival for \(EF \leq 30\) was lower than other two levels. Moreover, relatively small difference between the survival of patients with \(30<EF<45\) and \(EF\geq 45\) can be observed , the difference in the survival times was also statistically significant at 5% level (\(p<0.001\)).
  • There is significant differences in the survival times between patients with high blood pressure and those without (\(p=0.036\)) , patients with high blood pressure have lower survival time.

Cox proportional Hazard

Unadjusted
Adjusted
Characteristic N HR 95% CI p-value HR 95% CI p-value
platelets_cat 299
< Q1 — — — —
< Q2 0.70 0.43, 1.14 0.15 0.64 0.38, 1.08 0.091
< Q3 0.79 0.49, 1.27 0.3 0.78 0.47, 1.29 0.3
Gender 299
F — — — —
M 1.01 0.67, 1.54 >0.9 0.77 0.48, 1.26 0.3
Age 299 1.04 1.03, 1.06 <0.001 1.05 1.03, 1.07 <0.001
EFraction_cat 299
30-45 — — — —
EF<=30 3.18 2.03, 4.97 <0.001 3.46 2.16, 5.55 <0.001
EF>45 1.30 0.69, 2.44 0.4 1.22 0.64, 2.33 0.5
Smoking 299
No — — — —
Yes 0.99 0.64, 1.53 >0.9 1.09 0.67, 1.77 0.7
Sodium 299 0.93 0.90, 0.97 <0.001 0.93 0.89, 0.97 0.001
BP 299
No — — — —
Yes 1.55 1.03, 2.33 0.037 1.73 1.13, 2.66 0.012
Diabetes 299
No — — — —
Yes 0.96 0.64, 1.44 0.8 1.08 0.70, 1.65 0.7
Anaemia 299
No — — — —
Yes 1.40 0.94, 2.09 0.10 1.33 0.86, 2.06 0.2
CPK 299 1.00 1.00, 1.00 0.3 1.00 1.00, 1.00 0.006
Abbreviations: CI = Confidence Interval, HR = Hazard Ratio
  • The preliminary results here show that variables Age ,Ejection fraction, Sodium, blood pressure and CPK are statistically significant at 5% level adjusting for other predictors.
  • The fitted univariate models for the same variables indicate that they are statistically significant in predicting mortality for heart attack patients.

Optimal Cox proportional Hazard

  • The forest plot below shows the optimal model after performing a stepwise selection (sequential replacement).
  • It has both the least Akaike information Criterion and Bayesian information Criterion (\(957.72 ~and~ 975.67\)) as compared to the full model (\(964.11~and~994.88\)) indicating that it is a better fit.
  • The likelihood ratio test suggests that adding more variables to the reduced model does not improve the model efficiency (\(\chi^2 =3.612,p=0.6065\)).

Interpretations

  • With a unit increase in age, there is 5% significant increase in the hazard of death for patients with heart attack (\(p<0.001\)) .
  • hazard of death for patients with \(EF \leq 30\) was \(3.4\) folds significantly higher as compared to patients with ejection fraction between \(30 ~and~45\) (\(p<0.001\)). those with \(EF>45\) had \(21\%\) more risk of dying compared those in baseline category though not statistically significant.
  • A one unit \((meq/L)\) increase in Serum sodium significantly decreases the hazard of death by 6% when adjusting for other predictors (\(p-value = 0.003\)).
  • An Anemic patient had 41% more chances of dying as compared to a non-anemic patient when adjusting for other covariates , though not significant (\(p-value <0.05\)).
  • CPK was found to be statistically non-significant since hazard ratio was 1 , even though the \(p-value\) is less than 5%.
  • Patients with high Blood pressure have 76% increased hazard(risk) of dying as compared to those who do not have high blood pressure, and this is statistically significant at 5% level (\(p<0.009\)).

Multicollinearity

  • The plot below shows that all variables had variance inflation factor within the range of 1 to 2 indicating that multicollinearity is not a problem in our final model.

Proportional hazards assumption

  • The plots below show that the test of proportionality is not significant for each of the covariates except for Ejection fraction.
  • The global test is also not statistically significant implying that our proportional hazards assumption is reasonable.

Martingale and Deviance Residuals

  • The plots of residuals below are roughly symmetrically distributed about zero hence there is minimal signs of outliers.

Coxsnell residuals

  • The plot of coxsnell residuals below shows that our model is not a 100% perfect fit but the cumulative hazard of the residuals follows some degree of a 45° line (unit exponential distribution) to suggest an approximately perfect fit.

Discussion

The statistical analysis identified age, Ejection fraction, sodium and BP as the significant variables affecting the likelihood of mortality among heart failure patients. Both these variables were observed to be associated with an increased hazard of death. The findings that seem surprising are non-significance of smoking and diabetes. However, similar results concerning diabetes and smoking have been reported in other studies as well. The reason behind may be smoking and diabetes are basically causes of heart problem at initial stages. We were only concerned with patients of NYHA class III and IV which are advanced stages of heart failure. Up to these stages, these factors (diabetes and smoking) may probably be controlled by medications and hence these factors do not have significant effect on deaths due to heart failure in class III and IV.

Performance of model was checked using Akaike Information Criterion ,Bayesian Information Criterion and residuals such as coxsnell , deviance and martingale residuals and all these suggest that our final model fits the data pretty well.

References

  1. WHO. Fact sheet on CVDs. Global Hearts. World Health Organization. 2016.
  2. Al-Shifa IH. Cardiac Diseases in Pakistan [Internet]. 2016 . http://www.shifa.com.pk/chronic-disease-pakistan/
  3. Pillai Harikrishnan Sivadasan Ganapathi S. Heart failure in South Asia. Curr Cardiol Rev. 2013;9: 102–111. pmid:23597297.

Thank you!

https://bongani-ncube.netlify.app/