Modeling Patient Flow in Community Health Centers (Puskesmas)

1. Define the process map. Identify each service node (registration, triage, pharmacy, laboratory, doctor consultation, health education, exit).
2. Collect data. Measure arrival patterns (hourly, daily, seasonal), servicetime distributions, and routing probabilities.
3. Choose a modelling tool. Options include opensource libraries (SimPy, AnyLogic PLE) or spreadsheetbased simulators.
4. Build the simulation. Code each node as a resource with a capacity (number of staff) and attach the statistical distributions.
5. Validate the model. Compare simulated average waiting times with observed values; adjust distributions as needed.
6. Run experiments. Alter staffing levels, add rooms, or change appointment policies and record performance metrics.

Key Metrics to Track

Case Study: Improving Flow at a Rural Puskesmas

Background: A Puskesmas serving a catchment of 25000 residents observed a 30minute average waiting time for generalpractice consultations, with frequent peaks on Wednesdays (vaccination day).

Data Collected (2week sample)

• Arrival rate: 812 patients per hour, Poisson distribution.
• Registration time: Mean 2min, exponential.
• Triage time: Mean 3min, lognormal.
• Doctor consultation: Mean 7min, gamma distribution.
• Pharmacy dispensing: Mean 4min, exponential.
• Routing: 70% go to doctor, 20% to lab, 10% to healtheducation session.

Model Construction (SimPy snippet)

import simpyimport randomdef patient(env, name):    with reg.resource.request() as req:        yield req        yield env.timeout(random.expovariate(1/2))    with triage.resource.request() as req:        yield req        yield env.timeout(random.lognormvariate(1.1, 0.3))    # routing decision    if random.random() < 0.7:        # doctor path        with doctor.resource.request() as req:            yield req            yield env.timeout(random.gammavariate(2, 3.5))        with pharmacy.resource.request() as req:            yield req            yield env.timeout(random.expovariate(1/4))    elif random.random() < 0.2:     # lab path        # lab process         pass    else:        # healtheducation         pass    # patient exits    

Experiment Results

• Baseline (1 doctor, 1 registration clerk): Avg. wait 28min; doctor utilization 92%.
• Adding a second doctor reduces wait to 13min; utilization drops to 78%.
• Introducing an appointment slot for vaccination day cuts peak arrival rate by 35% and brings overall wait time to 10min.

The simulation indicated that a modest staffing increase (one extra doctor) combined with a simple appointment system would achieve the target waiting time without significant cost increase.

Conclusion

Modeling patient flow in Puskesmas provides a datadriven foundation for operational improvements. By capturing realworld arrival patterns, service times, and routing probabilities, administrators can test whatif scenarios safely, allocate resources more efficiently, and ultimately deliver faster, higherquality care to the community.

Key takeaways:

• Start with a clear process map and reliable data collection.
• Use discreteevent simulation to reflect the stochastic nature of healthservice demand.
• Focus on actionable metrics such as waiting time and resource utilization.
• Iterate: validate the model, run experiments, and implement the most effective changes.

With these steps, any Puskesmas can transform raw patient movement into strategic insight, strengthening primary health care across Indonesia.