Model integrated postpartum care

Postpartum women in many settings face two concurrent needs: contraception to space or limit births, and HIV prevention during a period of elevated exposure. A single postnatal visit can offer both. This how-to shows how to model a co-delivery intervention – long-acting injectable contraception plus long-acting PrEP – at the same 2-month postpartum contact, using FPsim for the family planning side and STIsim’s HIV module for the PrEP side.

You will need FPsim and STIsim installed.

Sim skeleton

FPmod exposes ti_live_birth on every woman, which is the timestep of her most recent live birth. That is the natural hook for a postpartum intervention: on any given timestep, the eligible cohort is the women whose ti_live_birth was exactly visit_month timesteps ago and who are not currently pregnant.

Wire FPsim, STIsim HIV, a structured sexual network, and a maternal network into one sim, then hand the postpartum cohort to a small custom intervention:

import starsim as ss
import stisim as sti
import fpsim as fp

def make_sim(seed):
    hiv = sti.HIV(init_prev=0.05, beta_m2f=0.012)
    intvs = [sti.HIVTest(test_prob_data=0.9), sti.ART(coverage=0.9),
             PostnatalPackage(offer_fp=True, offer_prep=True)]
    return fp.Sim(
        location='kenya', n_agents=20_000, start_year=2000, end_year=2035, rand_seed=seed,
        diseases=hiv,
        networks=[sti.StructuredSexual(), ss.MaternalNet()],
        interventions=intvs,
    )

The intervention

PostnatalPackage is a small custom ss.Intervention that identifies women at visit_month timesteps postpartum, draws attendance, and offers the two services to the same attendees:

class PostnatalPackage(ss.Intervention):
    """Offer long-acting injectable contraception and/or long-acting PrEP at a postnatal visit."""

    def __init__(self, pars=None, name='pp_shop', offer_fp=True, offer_prep=True, **kwargs):
        super().__init__(name=name)
        self.offer_fp = offer_fp
        self.offer_prep = offer_prep
        self.define_pars(
            visit_month=2,
            start_year=2027,
            p_attend=ss.bernoulli(p=0.5),
            p_fp_uptake=ss.bernoulli(p=0.3),
            p_prep_uptake=ss.bernoulli(p=0.3),
            prep_eff=0.95,
            dur_prep=ss.normal(ss.months(6), ss.months(1)),
        )
        self.update_pars(pars, **kwargs)

    def init_pre(self, sim):
        super().init_pre(sim)
        self.fp_mod = sim.demographics.fp
        self.cm = sim.connectors.contraception
        self.inj_idx = self.cm.methods['inj'].idx
        self.hiv = sim.diseases.get('hiv') if self.offer_prep else None

    def step(self):
        if self.sim.t.now('year') < self.pars.start_year:
            return
        fp_mod = self.fp_mod
        target_ti = self.sim.ti - self.pars.visit_month
        eligible = ((fp_mod.ti_live_birth == target_ti) & ~fp_mod.pregnant).uids
        attendees = self.pars.p_attend.filter(eligible)
        if not len(attendees):
            return

        if self.offer_fp:
            not_on_method = attendees[~fp_mod.on_contra[attendees]]
            starters = self.pars.p_fp_uptake.filter(not_on_method)
            fp_mod.on_contra[starters] = True
            fp_mod.method[starters] = self.inj_idx
            fp_mod.ti_contra[starters] = self.sim.ti + self.cm.set_dur_method(starters)

        if self.offer_prep:
            hiv_neg = attendees[~self.hiv.infected[attendees] & ~self.hiv.on_prep[attendees]]
            starters = self.pars.p_prep_uptake.filter(hiv_neg)
            dur = ss.years(self.pars.dur_prep.rvs(starters))
            self.hiv.start_prep(starters, eff=self.pars.prep_eff, dur=dur, adh=1.0)

Use ss.bernoulli and ss.normal for the draws (not numpy.random) so common-random- number pairing across scenarios stays intact.

What to look at

Run the sim once for each scenario you want to compare (typical grid: baseline, FP-only, PrEP-only, both) with several stochastic seeds, then extract:

• sim.results.fp.short_intervals – births at <24 mo interpregnancy interval.
• sim.results.hiv.new_infections_f – new HIV infections among women.

Both are annualized in the standard way (.annualize()) before you take differences against baseline.

Gotchas

• Visit at month 2, not month 1. FPmod sets ti_contra = ti + 1 at delivery, so the contraception connector re-chooses a method at month 1. If the intervention fires at month 1 the choice is overwritten immediately. Month 2 is safe.
• FPsim sexual activity and the STIsim sexual network are independent. A woman can conceive without a current partner in StructuredSexual, and vice versa. Acceptable for an illustrative composition; call it out if you use the pattern for anything quantitative.
• HIV calibration matters. sti.HIV needs realistic init_prev and beta_m2f for the setting, or the PrEP arm’s effect will be measured against an implausible baseline.

Full reference project

A calibrated Kenya version of this analysis lives at hiv_fp_kenya – the intervention above is demo.py:PostnatalPackage, and the full 4-scenario grid, calibrated HIV parameters, Kenya scale factor, and two-panel figure are all in the repo.