Derivation of a hierarchical risk structure within the INEEDHELP using classification and regression tree analysis
T Osotthanakorn, S SinphurmsukskulAbstract
Background
The INEEDHELP mnemonic identifies patients with advanced heart failure (HF) who may require referral for advanced therapies. Although the cumulative INEEDHELP burden is associated with adverse outcomes, whether a simple hierarchical structure within its components provides incremental prognostic value remains uncertain.
Objective
To derive and internally validate a hierarchical structure overlay within INEEDHELP and evaluate its incremental prognostic significance in patients with HFrEF.
Methods
In a retrospective cohort of 567 outpatients with HFrEF (2015–2024), INEEDHELP variables were used to construct a CHAID classification tree to identify hierarchical splits associated with a 1-year composite outcome (death, HF rehospitalization, LVAD implantation, or heart transplantation). The cohort was randomly divided into training and test samples using split-sample validation to assess structural reproducibility. The resulting structure identified edema/escalating diuretics (E) as the primary split, followed by NYHA class III/IV or elevated NT-proBNP (N). Terminal nodes were consolidated into three EN groups (Figure 1). Time-to-event analyses were performed using Kaplan–Meier methods and multivariable Cox models adjusting for age, sex, and total INEEDHELP count.
Results
Among 567 patients (mean age 61±17 years; 28.7% female), 1-year composite events occurred in 29.3%. The EN hierarchy demonstrated identical primary and secondary splits in both training and test samples, supporting internal reproducibility. Across the full cohort, EN groups showed significant survival separation (log-rank x² = 58.6, p < 0.001) (Figure 2). After adjustment for cumulative INEEDHELP burden, the EN overlay remained independently associated with outcomes (overall p = 0.006), with the highest EN category conferring a two-fold increased risk compared with the lowest category (adjusted HR 2.15, 95% CI 1.34–3.45; p = 0.002). Incorporation of EN significantly improved model fit beyond cumulative burden of INEEDHELP alone (Δ-2LL=9.92, p = 0.007).
Conclusion
A simple EN hierarchical overlay within INEEDHELP is internally reproducible and provides incremental prognostic stratification beyond the cumulative burden of INEEDHELP. This structure may add a benefit to prioritization among patients meeting INEEDHELP criteria.Figure 1 Classification treeFigure 2 Time to composite primary outcome