Demand Forecasting, Capacity Planning and Resource Allocation

Introduction

Hospitals must make decisions about beds, operating rooms, staff, diagnostic equipment, medicines and support services before the exact level of demand is known. Demand varies by hour, day, season, disease pattern, referral activity, emergencies and public-health events. Capacity that is too small produces queues, cancellations, overcrowding and unsafe workload. Capacity that is too large increases idle resources and operating cost. Hospital operations management therefore seeks a practical balance between access, quality, resilience and efficiency.

Demand forecasting estimates the volume and pattern of future service requirements. Capacity planning determines the resources required to meet that demand. Resource allocation assigns scarce resources among departments, patient groups and activities. These decisions are connected: a forecast of emergency admissions affects bed planning; bed availability affects elective scheduling; staffing levels affect the usable capacity of every clinical area.

This chapter develops the concepts and calculations needed by hospital managers to align capacity with demand while recognizing uncertainty, clinical priorities and ethical obligations.

Learning Objectives

By the end of this chapter, the learner should be able to:

1. Explain the relationship between demand, capacity, utilization and access.
2. Apply basic quantitative and qualitative forecasting methods.
3. Calculate bed requirements, occupancy, turnover and utilization measures.
4. Explain queueing concepts and identify hospital bottlenecks.
5. Allocate staff, beds, equipment and time using transparent priorities.
6. Interpret operational data and recognize the limitations of forecasting models.
7. Propose an integrated capacity plan for a hospital service.

1. Demand in Hospital Services

Demand is the quantity of care that patients seek or require during a specified period. It is not always identical to service volume. Some patients who need care may not present because of cost, distance or lack of awareness. Conversely, observed utilization may be influenced by referral policies, waiting lists or available capacity.

Important distinctions include:

• Need: the patient’s clinically appropriate requirement for care.
• Demand: expressed or anticipated requests for care.
• Utilization: services actually delivered.
• Unmet demand: required or requested care that is delayed, refused or inaccessible.
• Capacity: the maximum service output possible under defined conditions.
• Effective capacity: the realistically achievable output after considering staffing, maintenance, breaks, infection-control restrictions and variability.

Hospital demand has several characteristics. It is heterogeneous because patients differ in acuity and resource requirements. It is time-dependent because arrivals vary by hour and season. It is partly dependent: demand for laboratory tests, imaging and pharmacy services is generated by clinical encounters. It is also uncertain, particularly in emergency and disaster settings.

A useful operational classification is:

2. Demand Forecasting

Forecasting is a structured estimate of future demand based on historical data, clinical knowledge, external information and explicit assumptions. A forecast is not a promise. It is a decision-support estimate accompanied by uncertainty.

2.1 Forecasting process

A professional forecasting process normally includes:

1. Define the forecasting purpose, service, population and time horizon.
2. Select the measure, such as arrivals, admissions, procedures, patient-hours or occupied bed-days.
3. Collect and validate historical data.
4. Examine trend, seasonality, outliers and structural changes.
5. Select a suitable method.
6. Produce a central forecast and, where possible, a prediction interval or scenario range.
7. Compare forecasts with actual results.
8. Revise assumptions and communicate implications to decision-makers.

The time horizon matters. Short-term forecasts support daily staffing and bed management. Medium-term forecasts support rosters, equipment and clinic templates. Long-term forecasts support building, service-line and workforce decisions.

2.2 Basic forecasting methods

Naïve forecast. The next period is forecast as equal to the most recent actual value. This is simple and often useful as a baseline:

Here, F_t+1 is the forecast for the next period and A_t is the actual value in the current period. The method assumes that recent demand is the best available guide and is unsuitable when strong trend or seasonality exists.

Moving average. A moving average uses the mean of the most recent n observations:

All observations receive equal weight. It smooths random variation but responds slowly to sudden changes.

Weighted moving average. More recent observations may receive greater weights:

The weights must sum to one, making the forecast unit-consistent with the observations.

Exponential smoothing. A common form is:

where 0 < α ≤ 1 is the smoothing constant. A high α responds rapidly to recent changes; a low value produces greater smoothing. This method assumes that the recent level contains useful information and is most appropriate when trend and seasonal effects are limited unless extensions are used.

Trend and seasonal methods. A service with steadily increasing referrals may require a trend model. A service with predictable monthly patterns may require seasonal indices. Managers should avoid extrapolating historical patterns mechanically when there has been a change in clinical guidelines, payment policy, population structure or service configuration.

Judgmental forecasting. Clinicians and managers may provide expert estimates when data are limited or when a structural change makes historical data unreliable. Judgment should be documented, challenged constructively and combined with available evidence rather than treated as unquestionable opinion.

2.3 Forecast accuracy

Forecast errors should be monitored. For period t, the error can be defined as:

A positive error means actual demand exceeded the forecast. Common summary measures are:

MAE is expressed in the original units and is easy to interpret. RMSE penalizes large errors more heavily. MAPE is intuitive as a percentage but is unstable or undefined when actual demand is zero or very small. Forecast accuracy should be assessed against a simple baseline and interpreted alongside operational consequences. A small numerical error during a quiet period may matter less than a moderate error during a high-risk winter surge.

Worked Example 1: Moving-average forecast

A clinic recorded 120, 132 and 138 visits in the last three weeks. A three-week moving-average forecast for week four is:

The result is a forecast of 130 visits. It does not indicate that exactly 130 patients will arrive; staffing should consider the likely variation around this estimate.

3. Capacity Planning

Capacity is not simply the number of physical assets. Ten beds with insufficient nurses do not provide ten fully usable beds. Capacity depends on space, staff, equipment, operating hours, processes, supplies and regulatory constraints.

3.1 Capacity levels

• Design capacity: theoretical maximum under ideal conditions.
• Available capacity: capacity made available by planned operating hours and resources.
• Effective capacity: expected output after normal losses and constraints.
• Actual output: services delivered.

Capacity planning should distinguish volume capacity from flow capacity. Volume capacity concerns how much service can be stored or accommodated, such as beds. Flow capacity concerns how quickly patients can move through a process, such as imaging examinations per hour.

A basic utilization measure is:

The numerator and denominator must refer to the same service, period and unit. High utilization is not automatically good. At near-total utilization, even small fluctuations can create long queues and delays.

3.2 Bed capacity measures

The fundamental inpatient measure is the occupied bed-day. If a patient occupies one bed for one day, this represents one occupied bed-day. For a period:

If beds are temporarily closed, the number of operational beds—not the licensed total—should be used.

The occupancy rate is:

This calculation assumes that bed-days are measured consistently and that occupied and available beds relate to the same ward and period.

Average length of stay (ALOS) is commonly calculated as:

The denominator usually includes deaths and discharges, according to the organization’s reporting policy. ALOS is an average and should not be used alone to describe the distribution of stays.

Bed turnover is:

The turnover interval, sometimes called a bed turnover interval, estimates the average time a bed remains unoccupied between patients. A common form is:

Definitions vary across health systems, so local reporting standards must be followed.

Worked Example 2: Occupancy and bed requirement

A medical ward has 40 operational beds during a 30-day month. It records 1,020 occupied bed-days and 170 discharges.

Available bed-days:

Occupancy:

ALOS:

Bed turnover:

Turnover interval:

If projected monthly demand is 1,350 occupied bed-days and the planning target is 85% occupancy, required beds are:

The result should be rounded upward for planning, then tested against staffing, isolation needs, case mix and surge requirements. A mathematical result is not by itself an authorization to open beds.

3.3 Staffing capacity

Staff capacity should be expressed in productive hours, not merely headcount. If five nurses each provide seven productive hours per shift, the nominal capacity is 35 nurse-hours per shift. Breaks, handovers, training, leave, sickness and non-clinical work must be considered.

A workload-based staffing estimate can be represented as:

For example, if 80 outpatient visits are forecast and each requires 0.25 nurse-hours, required workload is 20 nurse-hours. If one nurse provides 6.5 productive hours, the calculated requirement is 20 ÷ 6.5 ≈ 3.08, so at least four nurses are required for coverage, subject to skill mix and safety rules. Standard times must be clinically credible and should not be used to pressure staff into unsafe performance.

4. Queueing and Access

A queue forms when demand temporarily exceeds service rate or when arrivals and service times vary. Hospital examples include emergency department waiting, imaging backlogs, outpatient appointment lists and discharge processing.

Key terms are:

• Arrival rate (λ): average arrivals per unit of time.
• Service rate (μ): average service completions per server per unit of time.
• Number of servers (s): parallel service channels, such as imaging rooms or registration desks.
• Utilization (ρ): the proportion of server capacity being used.

For a simple system with one server:

For ρ < 1, long-run stability is possible under the model assumptions. When ρ approaches 1, waiting time can increase sharply. If ρ ≥ 1, demand equals or exceeds service capacity and a persistent queue is expected unless demand is reduced or capacity is increased.

Little’s Law is a powerful relationship for a stable process:

Here, L is the average number of patients in the system, λ is the average throughput or arrival rate, and W is the average time in the system. Units must be consistent. If 12 patients per hour pass through a diagnostic process and average time in the system is 0.5 hour, the average number in the system is 12 × 0.5 = 6 patients.

Queueing models make assumptions about arrival patterns, service distributions, queue discipline and number of servers. Hospital systems often violate these assumptions because patients have different acuity, services cannot be interrupted, and clinical priority overrides first-come-first-served order. Queueing theory should therefore support—not replace—clinical judgment.

Managers can reduce waiting by:

• matching appointment supply to arrival patterns;
• separating urgent and routine streams where clinically appropriate;
• reducing unnecessary variation in service time;
• improving discharge and referral processes;
• adding flexible staff during predictable peaks;
• using triage and escalation rules;
• monitoring both average and extreme waiting times.

5. Bottleneck Analysis

A bottleneck is a resource or process step that limits total system throughput. It may be a physical constraint, such as one CT scanner, or a process constraint, such as delayed specialist review or lack of transport.

A practical bottleneck analysis follows these steps:

1. Map the patient journey from arrival to completion.
2. Measure demand, processing time, queue length and waiting at each step.
3. Identify the stage with persistent high utilization, backlog or longest delay.
4. Confirm that the apparent bottleneck is causal rather than merely visible.
5. Improve the constraint before expanding non-constraint resources.
6. Re-measure, because improving one constraint may shift the bottleneck elsewhere.

The Theory of Constraints provides a useful managerial logic: identify the constraint, exploit it using existing resources, subordinate other activities to it, elevate its capacity if necessary, and repeat the cycle.

For example, adding nurses to an inpatient ward may not improve admissions if the true bottleneck is delayed environmental cleaning or insufficient discharge coordination. Conversely, reducing average length of stay may not increase access if post-acute beds remain unavailable.

6. Resource Allocation

Resource allocation is the process of assigning limited resources among competing needs. Hospital allocation must combine efficiency with equity, clinical urgency, safety and legal requirements.

6.1 Allocation principles

A defensible allocation system should be:

• Clinically informed: based on acuity, risk and expected benefit.
• Transparent: criteria are documented and explainable.
• Consistent: similar cases are treated similarly.
• Responsive: decisions can change as conditions change.
• Equitable: it avoids discrimination based on irrelevant characteristics.
• Accountable: decisions can be reviewed and audited.

Resources may be allocated using several methods:

• Priority rules: emergency and time-critical patients receive precedence.
• First-come-first-served: suitable for some routine services but unsafe when urgency differs.
• Appointment templates: allocate time among urgent, routine, follow-up and procedural work.
• Skill-mix planning: assign tasks to appropriately qualified staff while preserving supervision.
• Optimization models: use constraints and objective functions to compare alternatives.
• Scenario allocation: prepare plans for normal, high-demand and crisis conditions.

An allocation objective may seek to maximize completed clinically appropriate activity, minimize waiting, reduce cancellations or maintain equitable access. These objectives can conflict. A hospital should explicitly state priorities rather than allowing hidden incentives to determine them.

6.2 Capacity buffers and escalation

Because demand and service times vary, operating at 100% planned utilization is unsafe for many services. A capacity buffer is deliberately reserved flexibility. Its size depends on variability, clinical risk, surge probability, staffing availability and the cost of excess capacity.

Escalation plans should define trigger points, actions, authority and communication. For example, a rising emergency department boarding count may trigger bed huddles, accelerated discharge review, conversion of designated space and executive escalation. Triggers should be based on reliable data and reviewed after each event.

7. Integrated Planning Framework

A hospital capacity plan should connect five questions:

1. What demand is expected? Forecast patient volumes, acuity and timing.
2. What capacity exists? Assess beds, staff, rooms, equipment, supplies and hours.
3. Where is variability greatest? Identify uncertain arrivals, long service times and surge risks.
4. What constraint governs flow? Locate bottlenecks and dependencies.
5. What action is proportionate? Adjust schedules, processes, staffing, referral patterns, capacity or contingency arrangements.

A monthly planning cycle may include a forecast meeting, a capacity review, a staffing and equipment check, a risk assessment, an agreed operating plan and a subsequent variance review. Daily operational control then adjusts the plan as actual demand becomes known.

Illustrative Case: Surgical and Medical Bed Capacity

Background

This illustrative case concerns a 120-bed district hospital. Elective surgery is frequently cancelled because emergency medical admissions occupy beds needed by postoperative patients. Managers initially proposed purchasing additional beds.

Problem

The hospital’s annual data show rising emergency admissions, variable elective demand and delayed discharges. The operating theatre is not fully utilized on some weekdays, while medical wards reach very high occupancy during winter. The hospital lacks a shared forecast and departments plan independently.

Analysis

The operations team separates demand into elective, emergency and seasonal components. It calculates occupied bed-days by service, daily occupancy rather than only monthly averages, average length of stay, discharge delays and theatre cancellations. Process mapping shows that the main constraint is not the number of physical beds alone. Delayed senior review, late prescriptions and limited transport prevent medically ready patients from leaving. These delays create a downstream bed bottleneck.

The team also identifies a mismatch between theatre scheduling and bed availability. Elective cases are booked without checking predicted postoperative bed demand. Emergency admissions peak on particular days, producing avoidable competition for beds.

Approach and solution

The hospital implements five linked actions:

1. A rolling four-week forecast combines emergency admissions, elective cases and seasonal indicators.
2. A daily bed-management meeting reviews expected admissions, discharges, isolation needs and staffing.
3. Discharge planning begins at admission, with early identification of transport, medicines and community-care needs.
4. Theatre templates reserve a proportion of capacity for urgent cases and align elective lists with forecast postoperative beds.
5. A winter escalation plan provides temporary staffing and a defined surge area when agreed thresholds are reached.

Forecasts are reported as ranges rather than single-point promises. Actual demand and forecast error are reviewed weekly.

Outcome

In the illustrative scenario, cancellations decline, discharge delays become visible by cause, and bed occupancy becomes less volatile. The hospital does not eliminate all peaks; instead, it improves coordination and uses flexible capacity where it has the greatest effect. The case demonstrates that capacity expansion without process improvement may simply increase cost while leaving the system constraint unchanged.

Learning points

• Bed capacity is a flow-system issue, not only a counting exercise.
• Forecasts must include timing and case mix, not just total volume.
• Bottleneck improvement often requires coordination across departments.
• Capacity plans should include uncertainty, buffers and escalation.
• Operational measures should be linked to patient access, safety and quality.

8. Governance, Ethics and Data Quality

Capacity decisions affect who receives care and when. Managers must ensure that allocation rules comply with applicable law, professional standards, emergency obligations, disability protections, privacy requirements and institutional policy. Personal health information should be protected, and analytical datasets should use appropriate access controls and de-identification where possible.

Ethically, efficiency cannot be the sole objective. A policy that reduces average waiting time but systematically disadvantages rural patients, people with disabilities or those with complex needs may be unacceptable. Equity indicators should be examined by relevant population groups, while avoiding inappropriate disclosure of small-cell information.

Data quality problems include duplicate records, inconsistent definitions of admission and discharge, missing timestamps, changes in coding practice and inaccurate bed-status information. Before using a dashboard, managers should define the data source, calculation method, update frequency and responsible owner.

9. Technology and Contemporary Developments

Electronic health records, real-time bed boards, appointment systems and workforce-management tools can improve visibility. Predictive analytics may identify likely admissions, deterioration or discharge delays. However, predictive models can reproduce historical inequities, fail during structural change and create false confidence. Human oversight, validation, monitoring for bias and clear accountability are essential.

Contemporary capacity planning increasingly emphasizes resilience. Hospitals plan not only for average demand but also for epidemics, cyber incidents, supply disruption, climate-related events and workforce shortages. Scenario planning, mutual-aid agreements, modular spaces and cross-training can improve adaptability. Digital systems are valuable only when the underlying processes, definitions and responsibilities are sound.

Best-Practice Checklist

A hospital manager should:

• define demand and capacity measures precisely;
• use operational rather than merely licensed bed counts;
• forecast by service, time period and acuity where possible;
• compare several methods and retain a baseline;
• report uncertainty and forecast error;
• monitor occupancy, throughput, waiting and cancellations together;
• identify bottlenecks through process observation and data;
• protect safety buffers and contingency capacity;
• align staffing, equipment, beds and schedules;
• document allocation criteria and escalation authority;
• review equity, privacy and data quality;
• evaluate outcomes and revise the plan.

Chapter Summary

Demand forecasting estimates future patient requirements using historical data, patterns, expert judgment and explicit assumptions. Naïve forecasts, moving averages, weighted averages and exponential smoothing are useful at different levels of complexity, but every forecast should be compared with actual results and interpreted with uncertainty.

Capacity includes physical space, staff, equipment, operating time and processes. Effective capacity is lower than theoretical design capacity because of normal operational losses. Bed planning requires consistent use of occupied bed-days, available bed-days, occupancy, average length of stay and turnover measures. Required beds depend on projected occupied bed-days and a chosen occupancy target, but the result must be tested against staffing, isolation and surge requirements.

Queueing concepts explain why waiting can rise sharply as utilization approaches full capacity. Little’s Law links average patients in a system, throughput and time in the system. Bottleneck analysis focuses attention on the constraint that limits total flow rather than on the most visible problem. Resource allocation should be clinically informed, transparent, equitable and adaptable.

Effective hospital operations integrate forecasting, capacity measurement, queue management, bottleneck improvement, staffing, scheduling and escalation. Technology can improve information and prediction, but governance, data quality, human judgment and ethical accountability remain essential.

Key Terms

Demand: The quantity of care requested or clinically required during a specified period.

Capacity: The maximum service output possible under defined conditions.

Effective capacity: Realistic capacity after normal operational losses and constraints.

Utilization: Actual output divided by available capacity.

Occupied bed-day: One inpatient bed occupied for one day.

Available bed-day: One operational bed available for one day.

Occupancy rate: Occupied bed-days divided by available bed-days, expressed as a percentage.

Average length of stay: Occupied bed-days divided by discharges.

Bed turnover: Discharges divided by average operational beds.

Forecast: An estimate of future demand based on evidence and assumptions.

Moving average: A forecast based on the mean of recent observations.

Exponential smoothing: A forecasting method that weights recent observations more heavily.

Forecast error: The difference between actual and forecast demand.

Queue: Patients or work awaiting service.

Arrival rate: Average number of arrivals per unit of time.

Service rate: Average number of services completed per server per unit of time.

Bottleneck: A constraint that limits the throughput of a system.

Capacity buffer: Deliberately reserved flexibility for variation or surge.

Resource allocation: Assignment of limited resources among competing needs.

Little’s Law: The relationship L = λW between average system content, throughput and time.

Review Questions

Short Answer Questions

1. Distinguish demand, need and utilization in hospital operations.
2. State two limitations of using occupancy rate as the only performance measure.
3. Define a bottleneck and give one hospital example.
4. What does the arrival rate represent in queueing theory?
5. Why should operational bed counts be used instead of only licensed bed counts?

Long Answer Questions

1. Explain the forecasting process and compare naïve, moving-average and exponential-smoothing methods.
2. Discuss the relationship between bed capacity, occupancy, average length of stay and bed turnover.
3. Explain why high utilization can produce disproportionately long queues.
4. Describe a systematic approach to identifying and managing a hospital bottleneck.
5. Discuss ethical and governance principles that should guide hospital resource allocation.

Analytical / Application Questions

1. A 30-bed ward records 765 occupied bed-days in a 30-day month. Calculate its occupancy rate and explain what additional information is needed before deciding whether capacity is adequate.
2. A diagnostic unit receives 10 patients per hour and the average time in the system is 36 minutes. Use Little’s Law to estimate the average number of patients in the system, showing consistent units.
3. A hospital experiences elective surgery cancellations despite having unused theatre hours. Develop an analysis that distinguishes whether the bottleneck is theatre capacity, postoperative beds, staffing, discharge flow or scheduling coordination.

Multiple Choice Questions with Answers

Question 1

A ward has 40 operational beds for 30 days and records 1,020 occupied bed-days. What is the occupancy rate?

A. 75%

B. 80%

C. 85%

D. 92%

Correct Answer: C. 85%

Question 2

Which measure is calculated as occupied bed-days divided by discharges?

A. Average length of stay

B. Bed turnover

C. Occupancy rate

D. Service utilization

Correct Answer: A. Average length of stay

Question 3

In a simple one-server queue, which condition generally supports long-run stability?

A. λ > μ

B. λ = μ

C. λ < μ

D. ρ > 1

Correct Answer: C. λ < μ

Question 4

A clinic recorded 120, 132 and 138 visits in three successive weeks. Using a three-week moving average, what is the forecast for the next week?

A. 126

B. 130

C. 132

D. 138

Correct Answer: B. 130

Question 5

Which statement best describes a bottleneck?

A. The least expensive process step

B. A resource that limits total system throughput

C. Any process with available spare capacity

D. A forecast with a large error

Correct Answer: B. A resource that limits total system throughput

Question 6

Little’s Law is represented by which relationship?

A. L = λW

B. L = λ ÷ W

C. W = L × λ

D. λ = L × W

Correct Answer: A. L = λW

Question 7

If 12 patients per hour pass through a process and average time in the system is 0.5 hour, what is the average number of patients in the system?

A. 6

B. 12

C. 24

D. 0.04

Correct Answer: A. 6

Question 8

Which forecasting accuracy measure is expressed in the original units of demand?

A. MAPE

B. MAE

C. Percentage bias only

D. Occupancy rate

Correct Answer: B. MAE

Question 9

Why is operating a hospital service at 100% planned utilization risky?

A. It guarantees lower quality costs

B. It eliminates all demand variation

C. Small fluctuations can create queues and delays

D. It always increases effective capacity

Correct Answer: C. Small fluctuations can create queues and delays

Question 10

Which is the most defensible principle for allocating scarce hospital resources?

A. Use undocumented personal preference

B. Give priority solely to the earliest arrival in every circumstance

C. Apply transparent, clinically informed and equitable criteria

D. Allocate resources only to the most profitable service

Correct Answer: C. Apply transparent, clinically informed and equitable criteria