Skip to content
Back to search results

Queuing Theory in Healthcare

Introduced by
Agner Krarup Erlang
Year
1909
Type
Stochastic modeling and optimization technique
Family
Operations research, Service management
Font size

What does it do?

Queuing theory is a mathematical discipline that models waiting lines, service capacity, and customer (patient) flow.

Developed initially by Agner Erlang for telecommunications in 1909, it has been extensively applied to healthcare to analyze and optimize emergency departments, outpatient clinics, surgical suites, and diagnostic service centers.

The intuition behind the method

Queuing problems arise when demand for service varies unpredictably and service capacity is fixed or limited. In healthcare, patients arrive at irregular intervals, their service times vary, and providers have capacity constraints. Queuing theory uses probability and statistics to predict queue lengths, wait times, and system utilization without needing to simulate every patient. By understanding these relationships, managers can decide how many staff members, beds, or equipment units are needed to achieve acceptable service levels.

History

Agner Erlang developed queuing theory in 1909 while analyzing telephone call volumes in Copenhagen. The field was formalized by David Kendall, who introduced the now-standard notation for queue types (M/M/1, etc.) in 1953. Healthcare applications accelerated in the 1990s and 2000s as operations research methods were applied to emergency department flow, surgical scheduling, and outpatient clinic management. Extensions to handle nonexponential distributions and time-varying arrivals have made the theory more applicable to real healthcare systems.

When to use it

Use queuing theory when analyzing service systems with variable demand and fixed capacity, such as emergency departments, outpatient clinics, and diagnostic services. It is effective for predicting system behavior without expensive simulation. Assume that arrivals and service times are random, that the system is in steady state (not in startup or shutdown), and that queue discipline is clear (FIFO, priority). Avoid queuing theory if arrival patterns are highly deterministic, if the system is very small (< 3 servers), or if non-standard queue disciplines make analysis intractable.

Limitations and considerations

  • Standard queuing models assume random (Poisson) arrivals and exponential service times, which often do not hold in healthcare
  • Does not account for patient abandonment, reneging, or complex routing patterns
  • Assumes stationary conditions; does not handle time-varying demand (e.g., seasonal variation)
  • Results are sensitive to assumption about distribution shape; small errors in assumptions can lead to large prediction errors
  • Limited ability to model interaction between multiple service stations or priority policies

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Provides closed-form formulas for waiting times, queue lengths, and utilization with minimal data
  • Allows rapid scenario analysis and 'what-if' studies (e.g., adding a provider reduces wait time by how much?)
  • Identifies bottlenecks and optimal staffing levels for given service targets
  • Well-established theory with extensive literature and software implementations
  • Low computational cost compared to simulation

Applications

  • Predicting emergency department wait times and determining optimal physician and nurse staffing for target service levels
  • Analyzing outpatient clinic capacity and deciding whether to add appointment slots, parallel clinics, or shift scheduling
  • Evaluating surgical suite utilization and predicting turnover times to maximize schedule efficiency
  • Assessing diagnostic imaging capacity (CT, MRI, ultrasound) and scheduling maintenance windows
  • Planning call center staffing for patient scheduling and advice lines based on expected call volume and target answer times

How it works

How it works

1

Collect data on patient arrival patterns (inter-arrival times), service times (how long treatment takes), and number of servers (physicians, nurses, beds, scanners). Data should span multiple weeks or months to capture variation and seasonality.

2

Summarize arrivals and service times as probability distributions (e.g., exponential, normal). Test whether observed data fit standard distributions using goodness-of-fit tests (Kolmogorov-Smirnov, Anderson-Darling).

3

Identify the queue structure: single server vs. multiple servers, single queue vs. multiple queues, and whether priority rules apply (emergency patients first, scheduled vs. walk-in).

4

Specify the queuing system using Kendall notation (e.g., M/M/c/N: Markovian arrivals, Markovian service, c servers, system capacity N) and calculate steady-state metrics: average queue length, average wait time, server utilization.

5

Determine whether current capacity meets service targets (e.g., 90% of patients seen within 4 hours). If not, identify whether to increase servers, reduce service time, or implement triage. Perform sensitivity analysis on key assumptions.

Common pitfalls

  • Assuming arrivals are Poisson when they are actually scheduled or have multiple peaks in the day
  • Ignoring patient heterogeneity (e.g., emergencies vs. routine cases) and applying standard queuing formulas to mixed systems
  • Confusing theoretical steady-state metrics with real-world first-day or worst-case scenarios
  • Using average service time when variability (standard deviation) is high; variability increases wait times much more than average alone suggests

Frequently asked questions

What is the M/M/1 queue and when is it used?

M/M/1 represents a single-server queue with Markovian (exponential) arrival and service time distributions. It is appropriate for modeling a single check-in clerk, a lone imaging scanner, or a single provider when demand is not too high. Its formula for average wait time is W = λ / (μ(μ − λ)), where λ is arrival rate and μ is service rate.

How does adding one more provider reduce wait time?

The effect is nonlinear and depends on current utilization. At low utilization (e.g., 30% server busy), adding capacity has minimal impact. At high utilization (e.g., 85% busy), adding one server can cut wait time dramatically. This is why healthcare systems often run near full capacity and see long waits.

Can I use queuing theory if patients arrive by appointment?

Partially. Scheduled arrivals reduce randomness and are more predictable, but appointment systems still have no-shows, cancellations, and variable service times. A hybrid model (some scheduled, some walk-in) is common in healthcare. If arrivals are highly deterministic (all appointments), simulation may be better.

What does 'utilization' mean and how does it affect wait time?

Utilization (ρ = λ / (c × μ)) is the fraction of server capacity being used. A 70% utilization means providers are busy 70% of the time. As utilization approaches 100%, wait times increase exponentially. Most healthcare systems aim for 70–85% utilization to balance efficiency and service levels.

How do I choose between adding staff vs. improving efficiency (reducing service time)?

Both reduce wait time, but efficiency improvements are often cheaper and more sustainable. Reducing average service time by 10% has similar effect to hiring more staff but without long-term salary costs. Prioritize process improvement first, then add capacity if needed.

Related methods

Sources

  1. Erlang, A. K. (1909). The theory of probabilities and telephone conversations. Nyt Tidsskrift for Matematik, 20(B), 33–39.

    Open publication reference
    Copied.
  2. Kendall, D. G. (1953). Stochastic processes occurring in the theory of queues and their application to the theory of failures. Annals of Mathematical Statistics, 24(3), 338–354.

    Open publication reference
    Copied.
  3. Gross, D., Shortle, J. F., Thompson, J. M., & Harris, C. M. (2008). Fundamentals of Queuing Theory (4th ed.). John Wiley & Sons.

    Open publication reference
    Copied.