---
title: Non-Homogeneous Poisson Process (NHPP) Framework
url: https://www.emergentmind.com/topics/non-homogeneous-poisson-process-nhpp-framework
type: topic
---

# Non-Homogeneous Poisson Process (NHPP) Framework

A non-homogeneous Poisson process (NHPP) framework provides a flexible, fully probabilistic paradigm for modeling and inference on recurrent event data where the conditional intensity function can vary with both time and covariates, capturing non-proportional, nonlinear, and potentially high-dimensional effects beyond the scope of classical survival analysis methods. Bayesian machine learning approaches, particularly Soft Bayesian Additive Regression Trees (SBART), have recently been developed to nonparametrically learn the NHPP intensity function in the presence of complex functional forms, subject-specific heterogeneity, and nonlinear covariate/time interactions [2606.12701].

## 1. NHPP Model Structure and Conditional Intensity Specification

The standard NHPP for recurrent events assumes that the counting process $N_i(t)$ for each subject $i\in\{1,\dots,n\}$ has conditional intensity
\[
\lambda_i(t \mid X_i, u_i) = \lambda_0 \cdot u_i \cdot g(X_i, t)
\]
where:
- $\lambda_0$ is a constant baseline hazard,
- $u_i$ is a latent subject-specific frailty ($u_i > 0$; typically Gamma-distributed for multiplicative random effects),
- $g(X_i, t)$ is an unknown function of covariates $X_i$ and time $t$ satisfying $g(X_i, t)>0$.

This formulation decouples baseline, subject-level, and time-varying/random effects, allowing $g(X_i, t)$ to model arbitrary smooth or complex interactions between the covariates and time. A common SBART representation is
\[
\lambda_i(t \mid W_i, X_i) = \lambda_0 W_i \Phi(b(t, X_i))
\]
with $b(\cdot)$ a latent regression function modeled by a sum-of-trees prior, and $\Phi$ the standard normal CDF so $g(X, t) = \Phi(b(t, X))$ [2606.12701].

## 2. Nonparametric Modeling of NHPP Intensity via SBART

Instead of imposing a fixed functional form for $g(\cdot,\cdot)$, SBART places an additive regression-tree prior on $b(t, x)$. Each tree $T_m$ partitions the input $(t, x)$ space via soft, probabilistic splits:
\[
T_m(x, t; \Psi_m, \mu_m) = \sum_{s=1}^{s_m} \omega_{ms}(x, t; \Psi_m) \mu_{ms}
\]
where $\Psi_m$ encodes the tree structure as a sequence of branch/leaf splits on joint time-covariate dimensions, and the soft path weights $\omega_{ms}$ are products of logistic (inverse-logit) gating functions. The full regression function is
\[
b(x, t) = \sum_{m=1}^M T_m(x, t)
\]
and the transformed intensity is $g(x, t) = \Phi(b(x, t))$ [2606.12701].

Key attributes of the SBART construction in this context:
- Supports high-dimensional, nonlinear, and potentially nonstationary intensity surfaces.
- Each branch node’s softness parameter $\alpha_b$ controls the transition sharpness; as $\alpha_b \rightarrow 0$, the model recovers hard partitioning.
- The prior over trees follows a Galton–Watson process with depth-dependent split probabilities, split variable/uniform cut allocations, and Gamma priors for bandwidth.

## 3. Hierarchical Priors and Latent Structure

The NHPP-SBART framework utilizes several hierarchically structured priors:
- Frailty effects: $W_i \sim \mathrm{Gamma}(\eta, \eta)$ with hyper-prior $\eta \sim \mathrm{Gamma}(a, b)$, enforcing $E[W_i]=1$ for identifiability.
- Baseline hazard: $\lambda_0 \sim \mathrm{Gamma}(\tilde{a}, \tilde{b})$.
- Tree structure: per-node split with probability $\gamma/(1+d)^\beta$ (standard choices: $\gamma = 0.95$, $\beta = 2$) [2606.12701].
- Leaf parameters: $\mu_{ms} \stackrel{\text{i.i.d.}}{\sim} N(0, \sigma_\mu^2)$, typically with $\sigma_\mu = 3/(2\sqrt{M})$ for $M$ trees.
- Bandwidths $\alpha_b$: shared per tree as $\alpha_m \sim \Gamma(1, r_\alpha)$ with $r_\alpha$ user-chosen.

## 4. Bayesian Computation: Data Augmentation and Posterior Inference

Direct likelihood evaluation in the NHPP framework involves terms of the form $\exp\{-\lambda_0 W_i \int_0^{a_i} g(X_i, t)dt\}$, which is analytically intractable with nonparametric $g$. SBART employs a two-layer data augmentation:

**Layer 1 (Poisson Thinning):** For each individual, simulate pseudo-events $\{G_{ik}\}$ from an independent NHPP with intensity
\[
\lambda_0 W_i [1 - g(X_i, t)]
\]
over $(0, a_i]$. The likelihood, conditioned on observed events $\{y_{ij}\}$ and pseudo-events $\{G_{ik}\}$, factorizes into products over $g(X_i, \cdot)$ and $1 - g(X_i, \cdot)$ [2606.12701].

**Layer 2 (Latent Gaussian Variables):** An Albert–Chib (probit) construction introduces latent $Z$: for each event $y_{ij}$, $Z_{ij} \sim N(b(y_{ij}, X_i), 1)$ truncated above $0$ (event), for each pseudo-event $G_{ik}$, $Z_{ik} \sim N(b(G_{ik}, X_i), 1)$ truncated below $0$ (no event). This converts the backfitting update for $b(\cdot)$ into a Gaussian regression problem, directly compatible with standard SBART Metropolis–Hastings moves [2606.12701].

Conditional posterior updates for frailties, baseline, and tree parameters are all available in closed form or via standard Gibbs/slice/MH steps.

## 5. Theoretical and Empirical Properties

The SBART-based NHPP approach provides the following properties:
- **Smoothness Adaptivity:** SBART can adapt to arbitrary (unknown) levels of smoothness in $g(\cdot,\cdot)$, with posterior contraction rates matching minimax optimality for Hölder classes [1707.09461].
- **High-dimensional/Interaction Recovery:** The additive sum-of-trees and Dirichlet splits induce automatic adaptation to sparsity, variable selection, and nonlinear interactions [1707.09461, 2210.16375].
- **Robustness under Misspecification:** Simulation studies confirm that SBART-based NHPP estimation yields lower mean squared error for cumulative intensity estimation and better goodness-of-fit via Martingale residuals even when frailty or intensity model assumptions are violated. For instance, in scenarios with $n=200$ subjects and $p=4$ covariates, RecSBART yielded average MSEs of $0.035$–$0.027$ across a range of simulated settings, outperforming both random forest and proportional-frailty models (random forest, $0.036$–$0.030$; proportional-frailty, $0.484$–$1.035$) [2606.12701].
- **Real-world validation:** Applied to colorectal-cancer recurrent hospitalization data ($n=403$), RecSBART achieved lowest mean squared Martingale residual (MSMR: $8.727$) versus RecForest ($10.153$) and a standard Bayesian proportional frailty model ($14.514$). Subject-level cross-validation showed RecSBART generalization gap $8.7\%$ vs. $28.4\%$ for RecForest [2606.12701].

## 6. Implementation, Scalability, and Practical Recommendations

The NHPP-SBART framework is implemented in compiled C++ code, enabling practical computation for moderate sample sizes (e.g., $n=200$ takes $\approx0.12$s per MCMC iteration). No additional computational accelerations beyond the two-layer augmentation are required for fit or mixing. Default choices for SBART hyperparameters (e.g., $M=50$, $\gamma=0.95$, $\beta=2$, $\sigma_\mu=3/(2\sqrt{M})$) are empirically robust. Posterior diagnostics rely on traceplots and effective-sample-size measures.

Empirical guidance for practitioners:
- Run sufficient MCMC iterations (e.g., $2,500$ burn-in$+2,500$ posterior draws).
- Sensitivity analyses on frailty priors exhibit <7\% effect on MSMR.
- Bayesian marginal effects (BMEs) can be quantified by integrating over covariate subspaces, and the proportional-intensity assumption is testable via examination of time-varying BME estimates; time-constancy in log-BME suggests proportionality [2606.12701].

## 7. Extensions and Applications

The NHPP framework with SBART generalizes to handle interval- and right-censored survival data, spatial clustering, and unobserved covariates by integrating further latent-variable schemes (e.g., spatial CAR priors for cluster frailties) and multi-stage data augmentation [2411.06591]. Smooth intensity estimation, uncertainty quantification, and predictive analysis are well supported. This approach is particularly suitable for biomedical and epidemiological recurrent event data, where subject-level heterogeneity and nonlinear dynamics cannot be parsimoniously specified in classical parametric or semiparametric survival models.

### Summary Table: NHPP–SBART Framework Components

| Component                | Specification/Role                               | Reference      |
|--------------------------|--------------------------------------------------|----------------|
| Intensity $\lambda_i(t)$ | $\lambda_0 W_i g(X_i, t)$                        | [2606.12701]   |
| $g(X_i, t)$              | SBART: $\Phi(b(x, t))$ sum-of-soft-trees         | [2606.12701]   |
| Frailty $W_i$            | $\mathrm{Gamma}(\eta, \eta)$, hyperprior on $\eta$| [2606.12701]   |
| Data augmentation        | Poisson thinning + latent Gaussian variables      | [2606.12701]   |
| Computation              | Tree-wise MCMC, conjugate updates, C++ impl.     | [2606.12701]   |
| Empirical validation     | Simulation, real-world hospitalizations          | [2606.12701]   |

The NHPP framework with SBART establishes a scalable, nonparametric statistical model for recurrent events. Each modeling and inferential step, including data augmentation, hierarchical prior specification, and adaptation to nonlinearity, is formally documented in the cited literature [2606.12701].

Source: https://www.emergentmind.com/topics/non-homogeneous-poisson-process-nhpp-framework