= Solution
The ordinary <Nelson–Aalen estimator> adds $d_j/r_j$ at event time $t_j$. Under a fitted <proportional hazards model>, replace the number at risk by the sum of fitted <hazard multipliers>:
$$
\widehat H_0(t)=\sum_{t_j\leq t}\frac{d_j}{\sum_{i\in R(t_j)}\widehat\phi_i}.
$$
This is the <Breslow estimator> of the integrated <baseline hazard>, using fitted <regression coefficients> obtained by maximizing the <Cox partial likelihood>. The denominator arises because the total event intensity is $h_0(t)\sum_{i\in R(t)}\phi_i$; with every multiplier equal to one it reduces to the <Nelson–Aalen estimator>.
The estimator is right-continuous, so $\widehat H_0(t_a)$ already includes the event at $t_a$. Subtracting it leaves only the events at $t_c$ and $t_d$, whose <risk sets> are $\{c,d\}$ and $\{d\}$. Hence
$$
\boxed{\widehat H_0(t_d)-\widehat H_0(t_a)=\frac1{\widehat\phi_c+\widehat\phi_d}+\frac1{\widehat\phi_d}.}
$$
There is no increment at the censoring time $t_b$. If the lower endpoint had instead been $t_a-$, an additional $1/(\widehat\phi_a+\widehat\phi_b+\widehat\phi_c+\widehat\phi_d)$ would be present. The final singleton event can increase the <baseline hazard> estimate even though its <partial likelihood> contribution is one.
Back to article page
