Projections of Public Spending on Pharmaceuticals: A Review of Methods

Abstract

Background

Forecasting future public pharmaceutical expenditure is a challenge for healthcare payers, particularly owing to the unpredictability of new market introductions and their economic impact. No best-practice forecasting methods have been established so far. The literature distinguishes between the top-down approach, based on historical trends, and the bottom-up approach, using a combination of historical and horizon scanning data. The objective of this review is to describe the methods for projections of pharmaceutical expenditure that apply the “bottom-up” approach and to synthesize the knowledge of their predictive accuracy.

Methods

Projections of public pharmaceutical expenditure applicable to Western economies including a comprehensive method description and published 2000–2024 were searched in scientific databases (MEDLINE, EMBASE, and EconLit) and in gray literature (websites of international health organizations and national healthcare authorities). The data sources, assumptions about the future market dynamics, analytical approaches, and the projection results are summarized.

Results

Twenty-four out of 3492 screened publications were included, associated with nine expenditure projection models. Four models were developed for all reimbursable drugs in the USA, the UK, the Stockholm region (Sweden), and seven European Union (EU) countries: France, Germany, Greece, Hungary, Poland, Portugal, and the UK, respectively. The other five models concerned specific groups of medicines: orphan drugs in Belgium, the Eurozone plus the UK, and Canada, respectively; psychotropic medications in the USA; and outpatient intravenous cancer medicines in the Province of Ontario (Canada). For trend analysis, drug coverage claims or sales data were used, applying linear and/or nonlinear models. The budget impact of new launches and patent expirations was estimated through (a form of) horizon scanning, i.e., a systematic monitoring of the pharmaceutical pipeline, with engagement of clinical expert judgment. Projections with a predictive time window greater than 3 years largely relied on previously observed trends to model new market introductions. Four models were validated through an ex post comparison of projected and observed expenditure. The absolute difference between the forecasted and actual percentual change in pharmaceutical expenditure was: 0.3% (“UK model”), 1.9% (“Stockholm model”), and 2% (nonfederal hospitals, “US model”). The “Ontario cancer drug model” overestimated the actual expenditure by 1%. Overall, the largest errors were attributable to new market launches and unforeseen policy reforms. Prediction accuracy decreased substantially for forecasts beyond 1 year in the future. For two not validated projections, a face validity check was feasible. One of the models forecasted a decrease in pharmaceutical expenditure from 2012 to 2016 in six European countries, contrasting with the currently available statistics. A 10-year projection of orphan drug expenditure underestimated the number of rare diseases treated in Europe by over 100%.

Conclusions

Published projections of national pharmaceutical expenditure are scarce and marked by significant methodological variability. Short-term forecasts based on high-quality historical data and rigorous horizon scanning tend to be more accurate than long-term forecasts built on theoretical assumptions. The combination of mathematical algorithms and expert judgment should be further explored, to increase the accuracy and efficiency of pharmaceutical expenditure projections.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40273-024-01465-w.

Key Points for Decision Makers

Forecasting future public pharmaceutical expenditure is a challenge for national healthcare payers, especially owing to rapidly changing drug market dynamics. No best-practice forecasting methods have been established so far.

This review summarizes published approaches to the prediction of pharmaceutical spending, demonstrating a significant methodological variability. Short-term forecasts built on high quality historical data and a systematic analysis of the pharmaceutical pipeline, i.e., horizon scanning of medicines, show acceptable accuracy.

Methodologies for pharmaceutical expenditure projections have to be further explored to better support the national resource planning, with particular attention to an optimal combination of mathematical algorithms and expert judgment.

Background

In Organisation for Economic Co-operation and Development (OECD) member countries, governments and compulsory insurance schemes are the main payers of medicines dispensed to patients via retail pharmacies and hospitals [1]. While expenditure on retail pharmaceuticals remained rather stable across Western European countries in the past decade (2011–2021), the costs of hospital medicines increased substantially, reaching an average annual growth rate of 4–8%, [1] or higher [2]. Such an increase has been associated with the launch of expensive treatments such as antivirals, antineoplastic, and anti-inflammatory drugs. The emerging gene and cell therapies introduce a new era of ultra-expensive individualized treatments, with per-patient treatment cost ranging from €200,000 to €2,000,000 [3]. The implications of these innovative medicines for the healthcare budget, as well as their sustained therapeutic value, is hard to estimate. Overall, there is increasing concern regarding the predictability and sustainability of public costs of pharmaceuticals.

Accurate forecasting of public expenditure on medicines has been a challenging task for healthcare payers, as many factors may influence them. The demand-side factors include demographic and epidemiologic situation, prescriber preferences, pricing and cost-containment policies of the health authorities; the supply-side factors are predominantly influenced by clinical pipeline dynamics, rate of success of pivotal trials, patent expirations, competition from generics and biosimilars, and the marketing strategies of pharmaceutical companies [4].

The theoretical literature distinguishes between “top-down” and “bottom-up” approaches to projections of healthcare expenditure [4]. The top-down method, mainly used for long-term projections (≥ 5 years), is based on historical trends in expenditure and has a limited ability to predict the short-term impact of unforeseen events [5–7]. Bottom-up models use a combination of historical trends and information about the anticipated market dynamics and are more suitable for short-term expenditure projections [8, 9].

Horizon scanning, i.e., the identification of emerging medicinal therapies through systematic monitoring of the pharmaceutical pipeline, has been introduced in some countries in the past two decades with the purpose of anticipating new market entries, especially innovative medicines addressing previously unmet medical needs [10]. According to an OECD survey performed in 2018, 77% of participating countries develop bottom-up pharmaceutical expenditure projections, while less than a half use horizon scanning to estimate the budget impact of potential new market launches [4]. A best practice for forecasting pharmaceutical spendings for the purpose of public resource planning has not been established [4].

The objective of this systematic literature review is to map the published methods for projections of national pharmaceutical expenditure that apply the bottom-up approach and synthesize the knowledge of their predictive accuracy.

Methods

Data Sources

To identify the relevant studies, we searched the MEDLINE, EMBASE, and EconLit databases. The search strategy was based on the combination of the following terms and their proximate notions: (1) expenditure, (2) pharmaceutical products, and (3) forecasting (Online Resource 1). A snowball strategy was applied to the retrieved publications. In addition, websites of international organizations, e.g., OECD, WHO, ISPOR, EUnetHTA, and HTAi, and of national healthcare authorities were explored, and selected national organizations were contacted for further inquiries (see Online Resource 2 for the full list of gray literature sources).

Study Selection

Criteria for inclusion in the review were a combination of the following study characteristics: (1) method description for projections of national pharmaceutical expenditure, or a validation of such method; (2) applicable to Western economies (Western Europe, USA, Canada, and Australia); (3) includes analysis of budgetary impact of new market entries; (4) provides a comprehensive method description (process, data sources, analytical algorithms, etc.); (5) published in English in 2000 or later (based on the OCDE survey, most Western countries introduced a pharmaceutical expenditure tracking system [needed for expenditure projections] in the past 25 years) [4].

Exclusion criteria were: (1) pharmaceutical expenditure projections based entirely on historical data, (2) distinct budget impact analysis of specific therapies, (3) budget projections of total healthcare expenditures, or (4) lacking a detailed method description (e.g., opinion papers, news, and editorials).

Data Extraction and Synthesis Method

The data extraction captured both general information about the models (e.g., origin, purpose, scope, time horizon, update frequency, and validation) and specific elements of the model structure essential for the reproducibility of the projections, e.g., use of historical data (data sources, baseline period, applied statistical models, etc.) and approach to develop assumptions about the future market dynamics (uptake of recently approved drugs, anticipated market launches and substitution of drugs on the market, loss of market exclusivity, introduction of generics and biosimilars, etc.) (see the data extraction form in Online Resource 3). For each model, the development and application context, the modeling approach, and the validation results (if available) were summarized.

Results

Search and Study Selection

The search yielded 3492 publications, 24 of which met the inclusion criteria (Fig. 1). The outreach to national health authorities did not add any additional papers.

Fig. 1.

Flow diagram of identification, screening and inclusion of studies in the systematic literature review. BIA budget impact analysis

The included publications [8, 9, 11–30] were associated with nine different models (Table 1). Four models projected pharmaceutical expenditure for all reimbursable drugs, in either a national or a regional context—USA, UK, and Stockholm region (Sweden)—or for a group of countries. The other five models developed projections for specific groups of medicines: orphan drugs, with three distinct projections for Belgium, the Eurozone plus the UK, and Canada; psychotropic medications in the USA; and outpatient intravenous cancer medicines in the Province of Ontario (Canada). The projection time horizon varied between 1 and 10 years. A (form of) validation is available for four of the six models: the US and the Stockholm models were validated in separate research papers. The accuracy of first-year predictions of the UK and the Ontario cancer drug models was discussed in the original papers.

Table 1.

Studies included in the literature review and scope of the associated pharmaceutical expenditure projection models

Included studies	Country/region	Type of drugs	Further referred to as	Time horizon	Performed by	Validation
Hoffman (2013) [11] Schumock (2014) [12] Schumock (2015) [13] Schumock (2016) [14] Schumock (2017) [15] Schumock (2018) [16] Schumock (2019) [17] Tichy (2020)[18] Tichy (2021) [19] Tichy (2022) [20] Tichy (2023) [21] Tichy (2024) [22]	USA	All	US model	1 year	Academic researchers	Hartke (2015) [23]
Wettermark (2010) [24]	Stockholm region (Sweden)	All	Stockholm model	2 years	Health authorities	Linner (2020) [25]
O’Neill (2013) [8]	The UK	All	UK model	4 years	Consulting company	Discussed in the original paper
Remuzat (2014) [26]	France, Germany, Greece, Hungary, Poland, Portugal, the UK	All	EU-7 model	5 years	Consulting company	none
Denis (2010) [27] Schey (2011) [28] Lech (2022) [29]	Belgium EU-17 plus the UK Canada	Orphan	Orphan drug models	6 years 10 years 5 years	Consulting companies	none
Hodgkin (2016) [30]	USA	Psychotropic	US psychotropic drug model	9 years	Academic researchers	none
Murray (2020) [9]	Ontario province (Canada)	Outpatient intravenous cancer drugs	Ontario cancer drug model	1 year	Health authorities	Discussed in the original paper

Key Characteristics of Included Studies

All models forecasted gross drug expenditure from the healthcare payer perspective, i.e., costs to the system before deduction of possible company rebates under the so-called managed entry agreements common in Europe [31]. The USA, Stockholm, and Ontario cancer drug models have existed for over 10 years and been updated annually. The latter two have been developed and maintained by the Swedish and Canadian regional health authorities, respectively, and form a part of the overall national drug management system. The other six models were a one-off exercise performed by a consulting company or an academic research group. All models considered future changes in supply-driven factors such as new branded medicines, and some also analyzed the demand-driven factors, e.g., demographic changes and policy measures (Table 2). Historical expenditure trends were analyzed by using national claims data, or IMS (currently known as IQVIA) drug sales data, [32] whereas the state-owned databases contained a higher level of detail, e.g., drug utilization per patient and per indication. The baseline period, i.e., the length of the expenditure history, varied from 1 to over 10 years (Table 3).

Table 2.

Factors influencing drug expenditures considered within the included models

Model factors	USA	Stockholm		UK	EU-7	US psychotropic drugs	Ontario cancer drugs	Orphan drugs
								Belgium	EU/UK	Canada
Supply-driven factors
New launch candidatesa	√		√	√	√	√	√	√	√	√
New generics	√		√	√	√	√	√	–	–	–
New biosimilars	√		√	√	√	√	√	–	–	–
Drugs losing market exclusivity	√		√	√	√	√	√	–	√	–
Uptake of recently reimbursed drugs	√		√	√	–	–	√	–	√	√
Demand-driven factors
Demographic changes	–		√	√	√	√	–	–	–	–
Changes in clinical guidelines	–		√	–	–	–	–	–	–	–
Policy measures	√		√	–	√	√	–	–	–	–

A tick mark (√) considered; a dash (–) not considered

^aDrugs with no historical data, i.e., recently approved drugs or anticipated for approval in the projection period

Table 3.

Type of historical data used in the models

Model	Data source	Baseline period	Statistical extrapolation method	Level of detail
				Product-level	Indication-level	Patient-level
USA	IMS Health/IQVIA National Sales Perspectives database	1–20a years	Linear and nonlinear models	Yes	No	No
Sweden	Sales data from the National Corporation of Swedish Pharmacies	4 years	Linear regression	Yes	Yes	Yes
U.K.	IMS’s British Pharmaceutical Index	2–8 years	Linear regression	Yes	No	No
EU-7	IMS database	1 year	Not reported	Yes	No	No
US psychotropic drugs	IMS pharmacy claims database	11 years	Not reported	Yes	No	No
Ontario cancer drugs	eClaims database	1 year	Linear and nonlinear models	Yes	Yes	Yes
Orphan drugs Eurozone/UK Canada Belgium	EFPIA and EMA Canadian Public Drug Claims Database IMS database, national pharmaceutical expenditure reports	10 years 11 years 9 years	N.A. Linear model N.A.	Yes Yes Yes	Yes Yes Yes	No No No

EFPIA European Federation of Pharmaceutical Industries and Associations [33], EMA European Medicines Agency [34], IQVIA formerly Quintiles and IMS Health, Inc., a multinational company serving the industries of health information technology and clinical research [32], N.A. not applicable

^aIMS/IQVIA data were available from the year 2000 onwards and used to present annual evolution in expenditures. However, the authors are not explicit about the baseline period used for annual projections

Model Description

The US Model

Since 1992, the American Journal of Health-System Pharmacy (AJHP) has been publishing annual reports describing trends for national prescription drug expenditures in the USA. Since 2000, this publication series has also included annual expenditure projections, complemented by a methodological section from 2013 onward [11–22]. The projections are prepared by independent researchers with the intent to help hospital and pharmacy managers develop clinical drug budgets.

Historical data for the trend analysis are obtained from the IQVIA National Sales Perspectives (NSP) database that contains all drug purchases for the entire US population. Various models are used: linear, exponential smoothing, autoregressive integrated moving average (ARIMA), and neural network models.

Factors anticipated to drive changes in pharmaceutical expenditure are classified as new products, price inflation, and volume and mix changes in utilization of existing products. Recently approved drugs are identified via the Food and Drug Administration (FDA) website, and drugs in the pipeline through IPD Analytics (brand and biosimilar pipeline database) [35].

Finally, drug expenditure projections for the next calendar year are generated by taking into account policy measures, new market launches, patent expirations, and other major factors believed to influence future drug expenditures. The anticipated drug expenditure growth is determined by the authors’ consensus.

In 2015, Hartke et al. evaluated the accuracy of the US model by comparing the AJHP forecasts with the actual drug expenditure in nonfederal hospitals and clinics1 [23]. The forecasting error was defined as absolute value of the difference between the forecasted and the actual percentage change in expenditures. The forecasts were directionally accurate (i.e., correctly forecasted increase or decrease in expenditure) in 3 out of 11 years and 6 out of 10 years for nonfederal hospitals and clinics, respectively. Actual growth was within the forecasted range in 2 out of 11 years for the nonfederal hospitals, and in 3 out of 10 years for the clinics. Over the years, the mean error for the nonfederal hospital and clinic drug expenditure forecasts were 2.0 (range 0.6–4.8) and 4.7 (range 0.7–12.0) percentage points, respectively. On the basis of these results, Hartke et al. qualify the US model forecasts as “reasonably accurate,” emphasizing that the dynamic nature of healthcare makes any prediction of expenditures unlikely to be perfect [23].

The Stockholm Model

The forecasting of drug utilization and expenditure was introduced in Sweden in 2003 and driven by the regional health authorities. The study by Wettermark et al. presents the approach to the pharmaceutical expenditure forecasting for the current and next year in the metropolitan area of Stockholm operated by the Regional Drug and Therapeutics Committee (DTC) [24]. The modeling approach includes the following steps: (1) a linear regression model is fit to a time series of drug utilization data from the previous 4 years per therapeutic group (TG)2; (2) a linear extrapolation is done for each TG, to produce “crude predictions” (i.e., projections based exclusively on historical trends); (3) adjustment of crude predictions for expected market dynamics including patent expirations, new launches, new guidelines from national bodies or the regional DTC, new policy measures, etc.; (4) review by clinical and pharmaceutical experts including joint workshops with expert groups; (5) model modification and refinement based on the expert input.

Horizon scanning, i.e., identification of emerging medicinal therapies through systematic monitoring of the pharmaceutical pipeline, is used to identify new substances and indications likely to be launched within 2 years. The Swedish Dental and Pharmaceutical Benefits Agency provides information on recently approved drugs. The potential impact on the healthcare budget is assessed based on estimates of the anticipated price for each new product, target patient populations, time for diffusion, and substitution effects. Prevalence and incidence data are collected from various sources including the Swedish National hospital discharge register, the National prescribed drug register, databases from the County Council, and published scientific studies. For drugs losing market exclusivity, expenditure reduction by 50–95% is assumed after introduction of generics, based on historical data. The impact of biosimilars is assessed individually.

A contribution from pharmaceutical companies is invited to estimate the impact of new launches. Furthermore, consultations are held with medical and scientific expert groups belonging to the DTC system and established per therapeutic area.

The Stockholm model was validated by Linner et al. [25]. This work compares the predicted pharmaceutical expenditure over 2007–2018 with the actual expenditure. The projections had a mean absolute error (defined as absolute difference between actual and predicted change in expenditure) of 1.9% over the assessed period (standard deviation, 1.3%). The forecasts for the same year were closer to the actual expenditure than the forecasts for the following year. The largest errors were observed for antivirals, immunosuppressants, antitumor necrosis factors (anti-TNFs), and monoclonal antibodies (mAbs). In 2013 and 2014, the respective over- and underestimations were more than 4%. The main reasons were misjudging the impact and timing of patent expirations (2013) and the speed of uptake of medicines for hepatitis C (2014).

The UK Model

The UK model was a one-off projection of the NHS pharmaceutical expenditure for the period 2012–2015 [8]. It was performed by a consulting company with the purpose to support the Association of the British Pharmaceutical Industry (ABPI) in preparing “The Pharmaceutical Price Regulation Scheme 2014,” a pact between the Department of Health and Social Care (DHSC) of the UK, NHS England, and ABPI that defined the affordable growth rate of pharmaceutical expenditures for the coming years.

The entire medicines market was disaggregated into four discrete categories: (1) new products to be launched in the coming 4 years; (2) products losing exclusivity in the coming 4 years (“LoE products”); (3) on-patent products launched in the previous 5 years; (4) older products (i.e., launched 5+ years prior to the year of analysis) not expected to lose exclusivity and generics. Projections for each product in each category were based on a combination of historical trend analysis and adjustments made using past experience with other medicines with a comparable market position (“drug analogues”). The analytical hierarchy was to (a) trend baseline data, (b) apply analogs, and (c) adjust in the light of expert input.

To project expenditure on planned launches (category 1), annual number of launches for each therapeutic area was estimated through analysis of the pharmaceutical pipeline and attrition rates of the clinical trials. The projected launch rates were compared with recent rates and validated by industry experts. To generate assumed disease-specific uptake curves3 and the (average) first year sales, IMS data were analyzed for launches in the previous 8 years (n = 367). Substitution rates were modeled based on historical analysis per therapeutic area; linear growth rate of old drug expenditure over the past 4 years was extrapolated as if new launches did not happen, “projected” sales were compared with actual sales, and the additional sales caused by new launches were classified as the marginal effect of new drugs. On the basis of these analyses and expert feedback, 25% of sales of future launches were assumed to be additive in the baseline scenario, with the exception of cancer drugs that tend to be used in combination with existing treatments or as third- or fourth-line treatments and where 75% of sales of future launches were assumed additive.

For loss of exclusivity (LoE) products (category 2), the observed impact of generic competition in the UK in recent years was used to project the impact of future patent expirations. Four types of LoE products were distinguished to construct respective price and volume erosion curves, depending on manufacturing complexity (“easy” or “difficult”) and channel (primary or secondary care). To model the future impact of LoE products, each medicine was assigned a LoE date and a price and volume erosion curve.

For recent brand medicines (category 3), all products were placed on relevant uptake curves using the results of the historical data analysis. For older drugs (category 4), projections were based on historical trends of the past 4 years, aggregated at ATC4 level.4

Pharmaceutical companies were consulted to validate all model assumptions. Model uncertainty was addressed by means of scenario analysis varying specific assumptions, e.g., number of launches by +/− 10%, higher and lower than average uptake/erosion curves, first-year sales by +/−10% compared with historical analogs, etc.

In a validation exercise (discussed in the same paper), the first-year projections (2012) were compared with the actual costs. For the secondary care drugs, the change in costs was underestimated by 3.8% (6.5% versus actual 10.3%); for the primary care drugs, it was overestimated by 1.8% (− 1.9% versus actual − 3.7%). Taken together, the forecast for the total pharmaceutical expenditure had a modest underestimation: 1% projected versus 1.3% actual growth. No analysis of the prediction accuracy beyond 2012 was performed.

The EU-7 Model

This one-off EU Pharmaceutical expenditure forecast was developed by a consulting company for the European Commission to inform the EU pharmaceutical strategy. Gross expenditures for all reimbursable medicines in France, Germany, Greece, Hungary, Poland, Portugal, and the UK were forecasted for 5 years (2012–2016) [36–38].

To estimate the change in pharmaceutical expenditure, the model considered the impact from LoE and introduction of generics/biosimilars (savings) and new market entries (additional costs). IMS sales data from the previous year (2011) were used as baseline data. Trends in population aging and changes in disease prevalence were factored in. Pharmaceuticals going off-patent and new branded medicines were identified through horizon scanning. Risk of trial failure was assessed by therapeutic area based on clinical trial information, e.g., initiation date, recruitment progress, and supported by the attrition rates published in literature. Estimated drug launches and marketing authorization (MA) dates were validated by experts. Efficacy and safety of new drugs compared with the available treatment alternatives were evaluated by using the Health Technology Assessment (HTA) perspective of each country.

A principal assumption was that sales of products with similar clinical value would replace sales of existing products and thus cause no budget impact. In the oncology area, the clinical impact assessment was based on the overall survival (OS) data. Products were considered as not assessable when the OS data were not available. Reimbursement was assumed to occur 1 year after MA in all countries except for Germany where approved drugs are available on the market immediately. Peak sales were assumed to be achieved 3 years after launch with a linear increase for all medicines except for orphan drugs where peak sales were expected 1 year after launch.

The analyses were performed for all drugs and subgroups of retail and hospital drugs and included a deterministic and probabilistic sensitivity analysis. The model predicted a gross drug expenditure reduction over 5 years in all countries, except for Poland. The highest reduction was predicted for the UK at € 9367 million and the lowest in Hungary at €84 million. No ex post model validation was performed.

The US Psychotropic Drug Model

This one-off expenditure projection for psychotropic medications used in mental health disorders such as depression, bipolar disorder, schizophrenia and ADHD, was developed for the period 2012 through 2020 for the US Department of Health and Human Services. The initiative was driven by the increasing costs of prescriptions for psychiatric drugs in previous years [30].

Three main potential factors of change in future drug spending were considered in the model: mix of prescription drugs, drug prices, and prescription volume. Horizon scanning was applied to identify possible market introductions, estimate the MA timing, and anticipate the comparative effectiveness of new medicines. Next, medications going off-patent were spotted. Possible changes in drug utilization were estimated through expert engagement. For generic versions of branded medicines, medication subclasses with a low share of generics (below 70%) were distinguished from those with a high share. For the group with a low share of generics, projections were made at the molecule level, i.e., the market shares and price after entry were estimated for both brand and generic versions of the same molecule. For the group with a high share of generics, projections were made at the therapeutic class level based on the expectation that any future brand patent expirations would affect only a minimal portion of these markets.

For the trend analysis, the IMS database of electronic pharmacy transactions over 2002–2012 was used. All medicines were disaggregated by patent status (brand, generic, or branded generic). Based on historical data analysis, literature review and expert advice, prices for generics were assumed to average 60% of the originator price during the year of introduction and 30% during the second year; generics were expected to capture 70% of prescriptions in the first year and 95% in the second year after market entry.

The base case scenario projections applied annual consumer price index (CPI) correction of 3.0–3.6%. Scenario analysis calculated effect of the Affordable Care Act (ACA) through estimating the take-up rates of this new insurance scheme and thus increase in the number of patients gaining coverage. The key assumptions were varied in turn by means of a deterministic sensitivity analysis, e.g., +/−10% change in speed of market entry and uptake, +/−1% annual price change of generics, etc.

The model estimated an increase in spending on psychotropic medications at an average rate of about 2.7% per year over the period from 2012 to 2020, a slow-down from the 7.5% average annual growth observed during the period 2002–2012. No ex post validation of these predictions was published.

The Ontario Cancer Drug Model

These projections were developed by the Provincial Drug Reimbursement Program (PDRP) at Cancer Care Ontario (CCO), which is responsible for monitoring and forecasting outpatient intravenous cancer drug spending in the province [9]. In 2018, over 40 injectable cancer drugs were funded through the PDRP, with approximately 150 corresponding reimbursable treatment pathways. While annual manual 3-year forecasting for each treatment pathway was implemented by CCO many years ago, this novel semi-automated model aimed at improving the usability and forecasting accuracy. The forecasts are used to assess risks of budget pressures and allow the Ministry of Health and Long-Term Care (MOHLTC) to inform drug policy decision-making.

In the first stage, automated forecasts were generated for each treatment pathway using linear and nonlinear forecasting models. In the second stage, forecasts were customized by expert users based on the sector knowledge. The “eClaims” database containing patient-level data on the total monthly doses administered under each treatment over the previous 12 months was used for historical trend analysis. The method included identification of “big budget drivers” (BBDs), i.e., treatments responsible for about 80% of the total budget and “small budget drivers” (SBDs), i.e., all other treatments. Distinct forecasting techniques were used for different drug subgroups: BBDs versus SBDs and based on the length of presence in the market, < 12 months versus 12–23 months versus ≥ 24 months. The subgroup analysis included stable BBDs (i.e., medicines with long time series and high expenditures not affected by major price/policy changes), unstable BBDs (i.e., medicines that were recently introduced and/or affected by external disruptions), and SBDs. The best models for each subgroup were selected based on financial forecasting literature and empirical validation [39–44]. The newly developed user interface allowed users to choose one of the programmed forecasting algorithms and customize the forecast. In the naive forecast (medicines with less than 12 months of claims data history), a customization of automated forecasts by experts was always required.

Ex post validation of the first-year forecast was performed through comparison of projected versus actual expenditures. The forecasts resulted in a 1% overestimation of expenditures, an equivalent of $Can 3 million (ca. €2.1 million). The authors report an improvement in forecasting accuracy compared with the previous CCO model and a time win for expert input due to partial automation. Stable BBDs were found best suited for an automated forecast. The forecast error for this subgroup was within ± 4%, where error is defined as the difference between forecasted and actual expenditure divided by the actual expenditure. The automated forecast for the unstable BBDs resulted in errors between 12 and 125%.

The Orphan Drug Models

Belgium The model was a single effort of a consulting company to calculate the impact of orphan drugs on the Belgian pharmaceutical budget in 2008 and to make an expenditure forecast over the following 5 years [27]. The IMS database and the national pharmaceutical expenditure reports were used for the trend analysis. The projections were based on three estimates: (a) the average annual number of orphan drugs that obtain MA in the EU, (b) the annual number of drugs that obtain a positive reimbursement decision in Belgium, and (c) the average annual per-patient cost of an orphan drug in Belgium. In the base case scenario, it was assumed that there would be an average increase in the number of MAs of ten drugs per year, and that 90% of orphan drugs with European MA would gain a national reimbursement. The average cost of a reimbursed orphan drug was estimated at €2.135 million. A scenario analysis explored the impact of lower and higher values of these assumptions on the projection results. The model forecasted an increase of the orphan drug budget from €66.2 million in 2008 to a value within the range between €130 and €204 million in 2013.

Eurozone and UK A single forecast of the ten-year evolution (2011–2020) of orphan drug turnovers in the Eurozone and the UK was performed by a consulting company [28]. For any given projection year, the total cost of orphan drugs was estimated according to the number of orphan diseases multiplied by the average annual cost of pharmaceutical treatment. The number of new orphan diseases was forecasted by using historical observations on the annual growth in the number of MAs for new orphan indications and assumed to remain constant over time. For the years 2001 to 2010, the number of orphan diseases treated in Europe was derived from the European Commission registry of orphan medicines with marketing approval. The annual treatment cost of an orphan drug was estimated based on the public prices and the treatment schemes described in the summaries of product characteristics.

The base case assumptions were: (a) the orphan drug designations rate after 2010 would grow at 10% per annum between 2011 and 2020, reflecting the trend between 2001 and 2010; (b) 43% of designated drugs would be for indications in which no other orphan drug was yet licensed; (c) the proportion of designated orphan drugs that were given MA would remain constant at 10.9% of all designations; (d) the average lag time between the orphan designation and MA was 4 years; (e) the number of treated patients for each specific indication was about 22% of the known prevalence; (f) in the tenth year after market introduction, the price would drop by 25% on average. The model forecasted around 110 rare diseases by 2020, with a median annual per-patient cost of €32,242 (varying between €1251 and €407,631 per patient per year).

Canada A 5-year projection of public spending (2021–2025) on nononcology orphan drugs in Canada was developed by a consulting company [29]. Total annual spending on all orphan and ultra-orphan medicines reimbursed between 2010 and 2020 was determined using the Canadian Public Drug Claims Database. Publicly available list prices and prevalence data were used to calculate average prevalence-weighted per-patient costs for orphan and ultra-orphan medicines. Future market launches were estimated through analysis of medicines with orphan designation from the FDA or EMA, active phase 3 clinical trials, medicines under regulatory or HTA review, and therapies in price negotiations with the drug plans. Future annual costs were estimated for each orphan medicine in the pipeline based on the historical prevalence-weighted per-patient cost and then discounted using various assumptions reflecting medicines attrition in the different steps between clinical trials and public reimbursement, as well as the estimated market penetration rate. Projected annual public spending on medicines to treat rare diseases was calculated by linearly extrapolating the historical costs of orphan drugs, and then adding the estimated annual costs of orphan drugs anticipated to be launched between 2021 and 2025. Scenario analyses included projection of historical costs alone, i.e., no adjustment for new market launches, adjustment for hypothetical rebates by pharmaceutical companies, alternative assumptions about the rates of positive HTA recommendations, and annual per-patient costs. The model estimated an increase in public spending on orphan drugs, from $380.9 million in 2020 to $1.6 billion in 2025.

No ex post validation was found for either of the orphan drug models.

Discussion

Published methods for the projection of national pharmaceutical expenditure are scarce. Our review identified nine bottom-up models marked by a high methodological variability, laying a foundation for further experimental work and scientific discussions.

Common Characteristics of the Models

While each model has its unique structure and analytic method, a number of common features could be identified:

• Use of historical data for trend analysis

• Use of (a form of) horizon scanning to identify planned launches and products going off-patent

• Clinical and pharmaceutical expert engagement for validation of the modeling assumptions

Methodological Variations

The analytical timespan ranged from 1 to 11 years. Though no consensus exists on the minimal baseline period needed for the projections, 4–5 years of history of pharmaceutical costs are likely to facilitate a trend analysis more adequately than 1-year history [8, 24].

The specificity of the data was more profound in the claims databases, which provide information on transactions at individual patient level, than in the commercial sales databases, which usually contained turnover information at product level. Since most medicines are marketed for multiple indications, patient-level expenditure per drug indication could improve the quality of expenditure monitoring and forecasts, as each indication is associated with a specific market position and dynamics.

There were different time gaps in availability of historical data on drug cost and utilization. In Sweden, for example, patient drug dispense information is available almost real time, allowing for an analysis of most recent trends, while other countries often have to deal with considerable delays.

The predictions were performed either in terms of drug expenditure or in terms of both expenditure and volumes. The analytical methods applied for the projection of historical expenditure data varied from simple linear extrapolation to more sophisticated predictive models such as ARIMA and neural network models. The predictive accuracy of specific models applied to different time series was not evaluated in the included studies.

While analytical forecasting methods based on the historical trends are reported in most papers, the approaches to the integration of information on the anticipated market dynamics remain ambiguous. One of the models predicted future costs by adding up the extrapolation of historical expenditure with the estimated budget impact of new launches [29]. As acknowledged by the authors, such approach may lead to overestimations. A consultation with the authors of the Stockholm model revealed that adjustments of the crude predictions are performed manually, based on annual discussions among the therapeutic area experts, and do not follow any mathematical algorithm or specific rules.

The models applied different forecasting horizons, from 1 to 9 years. While any time horizon may be justified and should be chosen based on the purpose of the projections, it should be noted that long-term bottom-up pharmaceutical expenditure projections are prone to very high uncertainties (see ‘Predictive accuracy’ and ‘Horizon scanning’ sections).

Predictive Accuracy

Published validation studies were available in only four models. In three of the four, the predictive error was defined as the absolute difference between the forecasted and actual percentual change in pharmaceutical expenditure. The error of 1-year projections was 0.3% in the one-off UK model, 1.9% in the Stockholm model (a mean over 12 validation years), and 2% for nonfederal hospital drugs in the US model (a mean over 11 validation years). The error was higher for certain drug subgroups, e.g., 3.8% for hospital drugs in the UK model, and 4.7% for the clinic drugs in the US model. The prediction error was defined differently in the Ontario cancer drug model, impeding the comparability of the results. However, overall, the largest errors were attributable to drugs with no or little historical data, deviations in timing of patent expirations, and unforeseen policy reforms. Analytical algorithms demonstrated limited value in prediction of costs associated with new or recent launches.

Validation of a forecast going beyond a 1-year horizon was performed only for the Stockholm model, reporting a substantial loss in the prediction accuracy for the next-year compared with the same-year forecast. The authors recommend an update of the forecast as close as possible prior to the decision date if forecasting is used to provide data for decisions on budget allocation and agreements between payers and providers [25].

For two out of the five projections without a published ex post verification of the results, a face validity check was feasible based on public sources. The EU-7 model forecasted a decrease in pharmaceutical expenditure in 2012–2016 in six European countries, [26] contrasting with the currently available statistics indicating a growth of the medicine bill over the forecast period in most countries [4]. Such underestimation may have been caused by a lack of attention to the uptake of recently approved drugs, and the adopted assumption that only drugs with added clinical value would impact sales. Especially (expensive) oncology drugs without comparative survival benefit have been frequently adopted as additive treatment or in the following lines of treatment, after failure of the existing drugs [8].

A 10-year projection of orphan drug expenditure calculated that about 110 rare diseases will be treated in Europe by 2020 [28]. In January 2024, the Orphanet database listed 6346 rare diseases, of which 4.84% (n = 307) are reported to benefit from an orphan medicinal product [45]. Hence, the model may have underestimated orphan drug spending by more than 100%.

Whereas most methods included horizon scanning for the identification of future launches, the orphan drug models adopted a more simplified approach to future market dynamics analysis, assuming a stable increase in the annual number of launches, [27, 28] a fixed median treatment cost per medicine, [27–29] and a fixed market penetration rate [28, 29]. A rapid development of the pharmaceutical pipeline and high costs of emerging technologies in recent years suggest that building the expenditure projections on assumptions derived from historical averages should be done with caution.

Horizon Scanning

A realistic forecast requires a horizon scanning of medicines to identify the anticipated market launches and estimate their potential budget impact. A preliminary evaluation of the therapeutic value, number of eligible patients, treatment cost, and launch time are part of the horizon scanning process, and are only possible once the results of the pivotal trial are known [46, 47]. It takes 2–3 years between the trial completion and reimbursed access of new therapies. With average success rate of pivotal trials lying below 60%, it is hard to estimate which products will actually enter the market 3+ years in the future [48].

Short-term forecasts that integrate horizon scanning and estimate a budget impact of each individual medicine (including patient population analysis, price, and substitution effects) appear to be more accurate than long-term forecasts. However, such analyses are resource intensive and require a well-functioning horizon scanning system [10]. In recent years, a European cooperation on the identification of emerging health technologies has been evolving with the implementation of the Beneluxa collaboration, [47] the International Horizon Scanning Initiative (IHSI), [49] and the joint work planned within the EU regulation on Health Technology Assessment [50]. Such collaboration should realize efficiency gains. Nevertheless, some elements needed for the estimation of the national budget impact remain country-specific, e.g., place in the treatment algorithm, number of patients eligible for treatment, and cost of current standard of care, and require a national horizon scanning capacity.

Study Limitations

The models included in our review do not necessarily reveal exhaustive evidence on the topic as some countries have been developing budget forecasts for internal use only [4]. Moreover, no systematic quality assessment of the included studies was possible, since pharmaceutical expenditure projection is an emerging research topic and no best-practice methods have been established yet.

Implications for Research and Practice

Despite the limited number of publications, our review identifies the essential elements of the projection methods that should be of interest for those involved in pharmaceutical resource planning, notably, healthcare payers at national or regional level.

General Practice Recommendations

• The purpose and policy application of expenditure projections should be clear, e.g., budgeting, preparation of policy reforms, estimation of changing expenditure dynamics within certain therapeutic areas, etc.

• The methods should support the predefined objectives and be well documented, allowing for reproducibility.

• A combination of spending trend analysis and horizon scanning information is essential for short-term projections to anticipate trend disruptions and areas of potential budgetary pressures.

• Availability of up-to-date data on drug expenditure for the trend analysis is essential. While some countries cope with time lags in the tracking of reimbursed medicines, a delay of even several months may negatively impact the quality of expenditure projections [24].

• Availability of expenditure data at indication level (rather than at substance level) may be preferable. Such a detailed approach to cost tracking should allow for an estimation of the market shares and differentiated price turnovers of new and existing medicines in specific therapeutic settings.

• Annual assessment of the accuracy of predictions and reasons for forecasting errors is important to improve projection methods [25]. Expenditure projections for budgeting purposes require high accuracy and have been updated by many countries several times per year [4]. However, it is worthwhile to note that predictions will never be perfect, especially if they are used to prepare healthcare policies such as cost containment measures.

Further Research

• An optimal approach to the historical data analysis remains ambiguous. Many European countries have been tracking their drug expenditure by collecting the data of ambulatory and hospital reimbursement claims at product (ATC-5) level [4]. Further research has to explore the predictive accuracy of different analytical models by using data with various baseline time windows and grouping drugs on the basis of the past expenditure dynamics. While for older drugs with a stable expenditure pattern linear models may be the best option, newly introduced expensive medicines may require more advanced flexible models [9].

• Methods to integrate horizon scanning forecasts into data-driven models need to be further explored. The integration of results from horizon scanning in a trend forecast is not straightforward. Simply adding the horizon scanning forecasts to the crude predictions could result in a major overestimation of the future expenditure because, through the use of historical data, the market introductions of the past are already reflected in the crude predictions, although it is unclear to which extent this covers the new introductions [29].

• Healthcare payers may be interested in methods to transversal budget impact of medicines, i.e., impact of the increasing pharmaceutical budgets on other healthcare services, e.g., possible curative effects of very expensive cell and gene therapies.

Conclusions

Validated methodologies for pharmaceutical expenditure projections are needed to better support national resource planning. Published projections of national pharmaceutical expenditure are scarce and marked by significant methodological variability. The dynamic nature of healthcare and the many factors that contribute to the increase or decrease of future spending, make any prediction of expenditures a complex analytical exercise. Short-term forecasts based on high-quality historical data and rigorous horizon scanning tend to be more accurate than long-term forecasts built on theoretical assumptions. The combination of mathematical algorithms and expert judgment should be further explored to increase the accuracy and efficiency of pharmaceutical expenditure projections.

Supplementary Information

Below is the link to the electronic supplementary material.

Declarations

Conflict of interest and Funding

All authors of the review are employed by the Belgian Health Care Knowledge Centre (KCE). KCE is a governmental institution, funded by the National Institute for Health and Disability Insurance (NIHDI), the Federal Public Service of Health, Food Chain Safety and Environment, and the Federal Public Service of Social Security. KCE’s mission is to advise policymakers on decisions relating to health care and health insurance on the basis of scientific and objective research. The authors have no competing interests.

Ethics approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.

Data availability

Not applicable.

Code availability

Not applicable.

Author contributions

All authors contributed to the study conception and design. Data collection and analysis were performed by I.O. and I.C. The first draft of the manuscript was written by I.O., and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.