Demand Forecasting and Supply Planning

Introduction

Healthcare material and equipment management is fundamentally a matching problem: the right item must be available in the right quantity, at the right place, at the right time, and at an acceptable cost. Demand is generated by outpatient visits, admissions, procedures, disease patterns, seasonal changes, emergencies, public-health programmes, and technological choices. Supply, however, is constrained by budgets, procurement rules, supplier capacity, transport, storage, shelf life, and regulatory requirements.

Demand forecasting converts evidence about future clinical activity into an estimate of required resources. Supply planning then translates that estimate into procurement, inventory, staffing, distribution, and budget decisions. Forecasting is not a promise of exact future demand. It is a disciplined estimate accompanied by assumptions, uncertainty, and a plan for monitoring and adjustment.

In healthcare, forecasting errors may have consequences beyond financial loss. A stock-out of oxygen, blood-collection tubes, insulin, or sterile surgical supplies can delay treatment and threaten life. Conversely, excessive procurement can cause expiry, obsolescence, tied-up capital, and unsafe storage conditions. The managerial objective is therefore not simply to minimize inventory, but to balance availability, resilience, affordability, quality, and equity.

Learning Objectives

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

1. Explain the purpose, scope, and limitations of demand forecasting in healthcare.
2. Select and apply qualitative, consumption-based, morbidity-based, and activity-based forecasting methods.
3. Analyse lead time, safety stock, reorder points, service levels, and forecast error.
4. Prepare a basic supply plan that links demand, inventory, procurement schedules, capacity, and budget.
5. Evaluate forecasting assumptions, data quality, ethical risks, and governance requirements.
6. Use scenario analysis and digital tools to improve planning without treating algorithmic outputs as unquestionable decisions.

1. Demand in the Healthcare Supply System

Demand is the quantity of a product or service that users require during a specified period. In healthcare, it is useful to distinguish three forms:

• Need: the amount required to achieve appropriate health outcomes, whether or not patients seek care.
• Utilisation: the amount actually used or consumed by the health system.
• Effective demand: the amount requested or purchased within available resources and service capacity.

These quantities may differ substantially. Low utilisation may reflect underdiagnosis, poor access, stock-outs, unaffordable fees, or weak reporting rather than low need. Historical consumption is therefore informative but not automatically an adequate forecast of future clinical requirement.

Demand may be classified as:

A forecast horizon is the period into the future covered by the forecast. Short horizons support replenishment; medium horizons support budgets and procurement cycles; long horizons support capacity, contracts, and capital investment. Forecasts normally become less accurate as the horizon increases.

2. Forecasting Architecture and Data Governance

A robust forecasting process begins with a clear planning question. “How many gloves will be needed?” is incomplete unless the organisation specifies location, glove type, clinical programme, time period, service standard, and whether wastage is included.

A practical forecasting architecture has six stages:

1. Define scope: products, facilities, population, time horizon, and planning unit.
2. Establish the baseline: clean historical consumption, activity, morbidity, stock, and procurement data.
3. Adjust for known changes: service expansion, formulary changes, outbreaks, substitution, price changes, and policy decisions.
4. Generate a forecast: apply an appropriate qualitative or quantitative method.
5. Convert forecast to supply plan: include opening stock, on-order stock, lead time, safety stock, pack sizes, and budget.
6. Review and learn: compare forecast with actual demand and document revisions.

Data quality should be assessed for completeness, accuracy, timeliness, consistency, and representativeness. Important data problems include duplicate records, unrecorded issues, transfers mistaken for consumption, retrospective adjustments, changes in units of measure, and stock-outs that make consumption appear artificially low.

A governance process should assign responsibility for data ownership, forecast approval, exception handling, and version control. Clinical, pharmacy, procurement, finance, epidemiology, and logistics personnel should participate because no single function sees the whole system.

3. Qualitative Forecasting

Qualitative methods are valuable when historical data are sparse, unstable, or structurally irrelevant. They are particularly useful for new hospitals, new products, outbreaks, technology transitions, and rare high-impact events.

Expert judgement and structured consensus

Experts estimate future demand based on clinical knowledge, policy, supplier information, and local experience. Unstructured judgement is vulnerable to anchoring, optimism, political pressure, and dominance by senior individuals. A structured process improves reliability by requiring written assumptions, independent estimates, evidence review, and explicit reconciliation of disagreement.

Delphi method

The Delphi method collects several rounds of anonymous expert estimates. A facilitator summarises the range and rationale after each round, allowing experts to revise their estimates. Anonymity reduces conformity and status effects. The method is useful for uncertain or emerging demand, but results remain judgemental and depend on the selection and expertise of the panel.

Scenario planning

Scenario planning does not seek one supposedly precise forecast. It constructs coherent alternatives, such as:

• Base case: expected activity and normal supply conditions.
• High-demand case: outbreak, referral increase, or accelerated programme expansion.
• Low-demand case: delayed implementation, reduced funding, or substitution.

For each scenario, managers identify trigger indicators, required stock, procurement actions, and financial consequences. Scenario planning is especially important for emergency commodities and long-lead equipment.

4. Quantitative Forecasting Methods

4.1 Naïve and moving-average methods

The naïve forecast uses the most recent observed value as the next forecast:

Here, F_t+1 is the forecast for period t+1, and A_t is actual demand in period t. It is a useful benchmark, not necessarily the best method.

A simple moving average of the last n observations is:

The method assumes that recent observations are representative and that trend and seasonality are limited. A longer window smooths random variation but responds slowly to change; a shorter window responds quickly but is more volatile.

Example. Monthly consumption was 980, 1,020, and 1,060 units. A three-month moving-average forecast is:

The unit is units per month. This forecast should be adjusted if a known service expansion will increase demand.

4.2 Weighted moving average

Recent periods may receive greater weights:

The weights are dimensionless and must sum to one. For demand values of 980, 1,020, and 1,060, with weights 0.2, 0.3, and 0.5 respectively, assigning the largest weight to the most recent value gives:

4.3 Exponential smoothing

Single exponential smoothing updates the previous forecast using the latest error:

α is the smoothing constant. A high value gives more weight to recent demand; a low value produces greater smoothing. This method is appropriate when demand has no strong trend or seasonality.

For a trend, Holt’s method may be used. For seasonal demand, a seasonal model or seasonal indices may be more appropriate. These models require sufficient, consistent data and should be validated rather than selected merely because they are mathematically sophisticated.

4.4 Regression and causal forecasting

Regression relates demand to explanatory variables such as admissions, operations, population, disease incidence, or clinic visits. A simple model is:

Here, y_t is demand, x_t is a driver, β₀ is the intercept, β₁ is the estimated change in demand for a one-unit change in the driver, and ε_t is unexplained error. The model assumes that the relationship is sufficiently stable and that the driver is measured reliably. Association does not prove causation, and extrapolation beyond observed data can be unsafe.

4.5 Forecast error and bias

For actual demand A_t and forecast F_t, define error as:

Common accuracy measures include:

MAE is expressed in the same units as demand. RMSE penalises large errors more heavily. MAPE is intuitive as a percentage but is unstable or undefined when actual demand is zero or close to zero. For intermittent healthcare demand, scaled errors, median absolute error, or service-level outcomes may be more meaningful.

Forecast bias can be monitored through mean error:

A positive mean error indicates systematic underforecasting under the definition above; a negative value indicates overforecasting. Accuracy should not be evaluated only at aggregate level: a good total may conceal dangerous errors for individual facilities or critical products.

5. Consumption, Morbidity, and Activity-Based Forecasting

Consumption method

The consumption method projects future requirements from historical issues or usage. A basic adjusted requirement is:

If observed consumption is distorted by stock-outs, the affected periods should be flagged or reconstructed. Adjustments may include programme growth, seasonality, substitution, wastage, and changes in pack size.

A useful supply requirement is:

All quantities must refer to the same product, unit, location, and time horizon. Expired, damaged, quarantined, or inaccessible stock should not be counted as usable stock.

Morbidity method

The morbidity method begins with the expected number of patients or cases and applies a standard treatment or diagnostic protocol:

If the population is 200,000, annual incidence is 2%, expected treatment coverage is 75%, and protocol use is 14 tablets per case:

This result excludes wastage, buffer stock, and opening inventory. It assumes the incidence estimate, coverage, and protocol are applicable to the population and period. Morbidity data can be biased by incomplete diagnosis, reporting delays, and access inequalities.

Activity-based method

For hospitals, supply requirements can be linked to planned activity:

For example, the number of surgical packs may be calculated from planned operations by procedure type. This method is strong when clinical protocols are standardised. It becomes unreliable when clinicians vary practice, activity plans are aspirational, or substitution is common.

Combining methods

A mature organisation triangulates methods. Consumption provides operational reality; morbidity estimates latent need; activity-based planning reflects service design; expert judgement addresses structural change. Large discrepancies should trigger investigation, not automatic averaging. The reason may be underuse, wastage, inappropriate prescribing, data error, or an unrealistic activity target.

6. Lead-Time Analysis and Inventory Protection

Lead time is the elapsed time between initiating an order and having usable stock available at the point of care. It may include requisition approval, tendering, supplier processing, manufacture, transport, customs, inspection, receipt, quality release, and internal distribution. The relevant measure is usable lead time, not merely supplier shipping time.

Lead time can be deterministic or variable. Managers should record both average lead time and its variability, and distinguish routine from emergency procurement. If demand during lead time is uncertain, inventory protection is required.

Reorder point

For constant demand and lead time:

d is average demand per time unit, L is lead time in the same time unit, and SS is safety stock. The first term covers expected lead-time demand; safety stock protects against uncertainty.

If demand is 100 units per week, lead time is 3 weeks, and safety stock is 80 units:

A replenishment order is triggered when inventory position reaches the reorder point. Inventory position equals on-hand stock plus on-order stock minus backorders, subject to the organisation’s inventory policy.

Safety stock and service level

If demand and lead time are approximately independent and demand variation is represented by standard deviation σ_d per period, a common approximation is:

Here, z is the standard-normal quantile associated with the desired cycle-service probability, and L is lead time in periods. The formula assumes stable demand, independent periods, approximately normal aggregate lead-time demand, and known lead time. It is unsuitable without modification for strongly intermittent, seasonal, censored, or highly skewed demand.

If weekly demand standard deviation is 20 units, lead time is 4 weeks, and z = 1.65:

If both demand and lead time vary, a commonly used approximation is:

and:

where σ_L is the standard deviation of lead time. If demand and lead time are correlated, a covariance term is required; the simplified formula should not be treated as universally valid.

A cycle service level is the probability of not experiencing a stock-out during a replenishment cycle. It differs from a fill rate, which is the proportion of demand supplied immediately from stock. A high cycle service level does not necessarily imply a high fill rate when shortages, order sizes, or demand variability are substantial.

Criticality should influence service targets. A lifesaving, non-substitutable item may justify a higher target and emergency reserve than a low-criticality administrative item. Service levels should therefore be clinically and financially governed, not selected solely by statistical convention.

7. From Forecast to Aggregate Supply Plan

Aggregate supply planning integrates forecasts across products, facilities, and time. It answers four questions:

1. What demand is expected?
2. What stock and supply are available?
3. What capacity and budget constraints apply?
4. What actions close the gap at acceptable risk?

A monthly planning table may include opening inventory, forecast demand, planned receipts, expected losses, closing inventory, purchase quantity, value, supplier, and risk status.

For a period-level stock balance:

The identity is a physical flow relationship. All terms must use the same unit and period. Negative closing stock indicates a shortage, not a valid inventory quantity.

Aggregate planning choices

Managers may use:

• Level supply: relatively stable procurement or production, with inventory absorbing variation.
• Chase supply: procurement or production changes with demand, reducing inventory but increasing capacity and supplier pressure.
• Hybrid planning: stable base supply plus flexible contingencies.

Healthcare generally requires a hybrid approach because routine demand can be levelled, while outbreaks, trauma, and seasonal peaks require flexible capacity. Aggregate plans should include supplier diversification, substitution rules, emergency procurement, allocation priorities, and redistribution between facilities.

Budget integration

A procurement budget should distinguish forecast quantity from authorised purchase quantity. A simplified purchase value is:

The total budget should also consider freight, customs, testing, installation, maintenance, training, disposal, and inventory carrying cost where relevant. Price assumptions should be explicit because exchange rates, inflation, and framework-contract conditions may alter affordability.

A plan that is clinically adequate but financially impossible is not an executable plan. Conversely, a budget-balanced plan that creates predictable stock-outs is not responsible management. Trade-offs should be documented and escalated when service targets cannot be met.

8. Technology, Analytics, and Artificial Intelligence

Enterprise resource-planning systems, electronic inventory records, barcode scanning, dashboards, and supplier portals can improve visibility and reduce transcription errors. Forecasting software can compare models, detect seasonality, and generate exception alerts. Interoperability is essential: a sophisticated model using incomplete or inconsistently coded data can produce false precision.

Artificial intelligence and machine learning may identify nonlinear relationships, estimate demand across many facilities, and detect unusual patterns. Their use requires sufficient historical data, stable definitions, monitoring for drift, explainability appropriate to the decision, cybersecurity, and human oversight. Models may reproduce inequities if historical utilisation reflects unequal access rather than true need. AI should support, not replace, clinical and managerial accountability.

9. Illustrative Case: Forecasting Essential Medicines During Service Expansion

Background

A district hospital plans to expand its diabetes clinic from 1,200 to 1,500 patient visits per month. Metformin tablets are issued in packs of 1,000 tablets. Historical usable consumption averaged 36,000 tablets per month, but one of the previous six months experienced a stock-out. The stock-out month recorded only 27,000 tablets issued. Supplier lead time is two months. Management wants one month of safety stock and expects 5% wastage.

Problem

A purely historical average would understate true demand because one observation was censored by the stock-out. The hospital must estimate demand, determine the procurement requirement, and avoid both interruption and excessive expiry.

Analysis

The five unaffected months averaged 36,000 tablets. The planned activity increase is:

Thus, activity is expected to be 25% above the historical clinic level. Adjusted usable demand is:

Expected wastage at 5% of usable demand is:

For a two-month lead time and one month of safety stock, the target stock requirement before the next replenishment arrives is:

Assume usable opening stock is 52,000 tablets and no stock is on order. The immediate gross purchase requirement is:

Because the product is purchased in packs of 1,000, the order must be rounded up to 90 packs, or 90,000 tablets. The resulting projected balance is 142,000 tablets, 250 tablets above the target.

Approach and solution

The hospital should document the censored stock-out, use the activity-based adjustment, include wastage explicitly, and place a 90-pack order subject to budget and expiry checks. It should also establish a monthly review of patient visits, issues, stock-outs, expiry, and supplier performance. If the clinic expansion is uncertain, a base and high-demand scenario should be maintained.

Outcome

The plan aligns procurement with expected service activity rather than distorted historical issues. It protects approximately one month of demand while avoiding an unnecessarily large order. The small rounding excess results from the pack-size constraint.

Learning points

1. Consumption data must be interpreted in light of availability; stock-out months are not normal demand observations.
2. Forecasts should reflect planned service changes.
3. Wastage and pack-size constraints affect the executable order.
4. A safety stock policy must be connected to clinical criticality and lead-time risk.
5. Forecast review is a continuing control process, not a once-yearly exercise.

10. Challenges, Ethics, and Best Practice

Major challenges include poor data quality, changing clinical protocols, supplier unreliability, fragmented budgets, demand surges, product substitution, expiry, and incentives that reward low expenditure rather than reliable care. Managers should avoid manipulating forecasts to obtain larger budgets or to conceal poor performance.

Equity is a central ethical issue. Allocation based only on historical consumption may perpetuate under-supply to remote or disadvantaged populations. Forecasts should consider unmet need, access barriers, and public-health priorities. Scarce supplies should be allocated using transparent, clinically defensible criteria rather than informal influence.

Regulatory and governance requirements commonly include authorised procurement, product quality assurance, cold-chain control, controlled-substance rules, traceability, record retention, segregation of duties, and protection of patient information. Forecasting datasets should be de-identified where individual-level data are unnecessary.

Best practice includes:

• segmenting items by clinical criticality, value, variability, and shelf life;
• combining consumption, morbidity, and activity evidence;
• documenting assumptions and data exclusions;
• measuring both forecast error and service outcomes;
• using rolling forecasts with formal monthly or quarterly review;
• setting differentiated service levels;
• modelling lead-time variability and supplier risk;
• involving clinicians, finance, procurement, and logistics;
• planning redistribution and substitution before shortages occur;
• validating automated recommendations through accountable human review.

Chapter Summary

Demand forecasting estimates future healthcare requirements under uncertainty, while supply planning converts those estimates into stock, procurement, capacity, and budget decisions. Effective planning distinguishes need, utilisation, and effective demand and recognises that historical consumption may be suppressed by stock-outs or access barriers.

Qualitative methods are valuable for new, uncertain, or rapidly changing situations. Quantitative methods include naïve forecasts, moving averages, exponential smoothing, regression, consumption-based calculations, morbidity methods, and activity-based planning. Forecast quality should be assessed with appropriate error measures and bias monitoring rather than a single accuracy statistic.

Lead-time analysis links demand during replenishment to reorder points and safety stock. Service level, fill rate, clinical criticality, and financial constraints must be considered together. Aggregate supply planning balances level, chase, and hybrid strategies and integrates inventory balances, pack sizes, wastage, supplier capacity, and budget. Digital systems and AI can improve visibility and analysis, but data quality, equity, explainability, cybersecurity, and human accountability remain essential.

Key Terms

Demand forecast: An estimate of future requirements for a specified period and scope.

Need: Health requirement that exists whether or not care is accessed.

Utilisation: Actual use of healthcare services or materials.

Effective demand: Demand expressed within available resources and capacity.

Forecast horizon: Future period covered by a forecast.

Consumption method: Forecasting based primarily on historical usage or issues.

Morbidity method: Forecasting based on disease incidence, population, coverage, and treatment requirements.

Activity-based forecasting: Estimating materials from planned clinical activities and standard quantities.

Lead time: Time from order initiation until usable stock is available.

Reorder point: Inventory position at which replenishment should be initiated.

Safety stock: Additional inventory held to protect against uncertainty.

Cycle service level: Probability of avoiding a stock-out during a replenishment cycle.

Fill rate: Proportion of demand supplied immediately from available stock.

Forecast error: Difference between actual demand and forecast demand.

Forecast bias: Systematic tendency to overforecast or underforecast.

Aggregate supply planning: Coordinated planning of supply, inventory, capacity, and budget across products or facilities.

Censored demand: Demand hidden by stock-outs or other supply constraints.

Scenario planning: Examination of alternative plausible future conditions and responses.

Inventory position: On-hand stock plus on-order stock minus backorders.

Service level: Target probability or performance standard for product availability.

Review Questions

Short Answer Questions

1. Distinguish need, utilisation, and effective demand in healthcare.
2. Why can historical consumption underestimate future requirement?
3. Define lead time and identify its major components.
4. What is the difference between cycle service level and fill rate?
5. State two limitations of MAPE in healthcare forecasting.

Long Answer Questions

1. Compare consumption-based, morbidity-based, and activity-based forecasting, including their assumptions and appropriate applications.
2. Explain how a healthcare organisation should design a governed forecasting process from data preparation through review.
3. Discuss the role of safety stock, reorder points, and lead-time variability in maintaining clinical availability.
4. Evaluate the usefulness and limitations of qualitative forecasting methods in a new hospital programme.
5. Explain how aggregate supply planning integrates demand, inventory, procurement capacity, and budget constraints.

Analytical / Application Questions

1. A clinic records monthly demand of 700, 760, and 840 units. Calculate the three-month moving-average forecast and explain one reason why the result may need adjustment.
2. A medicine has average weekly demand of 120 units, a lead time of 3 weeks, and safety stock of 90 units. Calculate the reorder point and state the assumptions behind the calculation.
3. A rural facility reports low consumption of a lifesaving product but frequent patient referrals caused by stock-outs. Which forecasting method or combination of methods would you recommend, and why?

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes effective demand from need?

A. Effective demand is always greater than need.

B. Effective demand is the requirement expressed within available resources and service capacity.

C. Effective demand excludes all clinical requirements.

D. Effective demand is identical to disease incidence.

Correct Answer: B. Effective demand is the requirement expressed within available resources and service capacity.

Question 2

A three-month moving average uses demands of 900, 1,000, and 1,100 units. What is the forecast for the next month?

A. 950 units

B. 1,000 units

C. 1,050 units

D. 1,100 units

Correct Answer: B. 1,000 units

Question 3

In exponential smoothing, increasing α generally makes the forecast:

A. Less responsive to recent observations

B. More responsive to recent observations

C. Independent of actual demand

D. Equal to the annual budget

Correct Answer: B. More responsive to recent observations

Question 4

A morbidity forecast normally combines disease incidence with which additional information?

A. Treatment coverage and quantity per case

B. Supplier profit and warehouse rent only

C. Last year’s purchase price only

D. The number of procurement officers

Correct Answer: A. Treatment coverage and quantity per case

Question 5

Average demand is 80 units per week, lead time is 4 weeks, and safety stock is 50 units. What is the reorder point?

A. 130 units

B. 270 units

C. 320 units

D. 370 units

Correct Answer: D. 370 units

Question 6

Which item should generally receive the highest service-level target?

A. A noncritical office stationery item

B. A lifesaving, non-substitutable medicine

C. An item with no clinical use

D. An obsolete product awaiting disposal

Correct Answer: B. A lifesaving, non-substitutable medicine

Question 7

What does a positive forecast error mean when error is defined as actual demand minus forecast demand?

A. Demand was lower than forecast

B. Demand equalled forecast

C. Demand was higher than forecast

D. The forecast was cancelled

Correct Answer: C. Demand was higher than forecast

Question 8

Which is a major limitation of MAPE?

A. It cannot be expressed as a percentage

B. It is unstable when actual demand is zero or near zero

C. It always penalises large errors more than RMSE

D. It requires no actual demand data

Correct Answer: B. It is unstable when actual demand is zero or near zero

Question 9

Why should a stock-out month often be excluded or adjusted in a consumption forecast?

A. It may record constrained issues rather than true clinical demand

B. It always contains duplicate invoices

C. It necessarily reflects peak demand

D. It makes all future forecasts invalid

Correct Answer: A. It may record constrained issues rather than true clinical demand

Question 10

Which approach best describes a hybrid aggregate supply plan?

A. No safety stock under any circumstances

B. Stable base supply combined with flexible responses to demand variation

C. Procurement based only on supplier preference

D. A plan that ignores budget constraints

Correct Answer: B. Stable base supply combined with flexible responses to demand variation