ASSET (Age/Sex Standardised Estimates of Treatment): a research model to improve the governance of prescribing funds in Italy.
ABSTRACT BACKGROUND: The primary objective of this study was to make the first step in the modelling of pharmaceutical demand in Italy, by deriving a weighted capitation model to account for demographic differences among general practices. The experimental model was called ASSET (Age/Sex Standardised Estimates of Treatment). METHODS AND MAJOR FINDINGS: Individual prescription costs and demographic data referred to 3,175,691 Italian subjects and were collected directly from three Regional Health Authorities over the 12-month period between October 2004 and September 2005. The mean annual prescription cost per individual was similar for males (196.13 euro) and females (195.12 euro). After 65 years of age, the mean prescribing costs for males were significantly higher than females. On average, costs for a 75-year-old subject would be 12 times the costs for a 25-34 year-old subject if male, 8 times if female. Subjects over 65 years of age (22% of total population) accounted for 56% of total prescribing costs. The weightings explained approximately 90% of the evolution of total prescribing costs, in spite of the pricing and reimbursement turbulences affecting Italy in the 2000-2005 period. The ASSET weightings were able to explain only about 25% of the variation in prescribing costs among individuals. CONCLUSIONS: If mainly idiosyncratic prescribing by general practitioners causes the unexplained variations, the introduction of capitation-based budgets would gradually move practices with high prescribing costs towards the national average. It is also possible, though, that the unexplained individual variation in prescribing costs is the result of differences in the clinical characteristics or socio-economic conditions of practice populations. If this is the case, capitation-based budgets may lead to unfair distribution of resources. The ASSET age/sex weightings should be used as a guide, not as the ultimate determinant, for an equitable allocation of prescribing resources to regional authorities and general practices.
- SourceAvailable from: Irene Petersen
- [Show abstract] [Hide abstract]
ABSTRACT: A longitudinal, retrospective, cohort study was designed to assess the mean treatment cost of invasive cervical cancer, cross-linking clinical and administrative databases. The study was performed from the Italian NHS perspective. Costs were estimated using DRG charges. All patients with histological diagnosis of invasive cervical cancer admitted to Gynecologic Oncology centres of the Catholic University of the Sacred Heart between 2000 and 2007 were enrolled. Overall, 351 patients (212 with LACC and 139 with ECC) were eligible for analysis. The mean direct medical cost by patient was € 22,200 ± 21,600 with a significantly higher burden for LACC compared with ECC patients (€ 28,696 ± 24,874 versus € 12,329 ± 8,726). Radical surgery accounted for € 6,851 ± 1,406, and € 7,709 ± 3,710 in ECC and LACC group, respectively; the resulting difference achieved a statistical significance. The extent of disease, disease progression/recurrence, and length of hospitalization confirmed their independent role as cost predictive factors. The combined use of administrative and clinical databases allowed a feasible assessment of the mean cost induced by the invasive cervical cancer. Management of LACC patients was associated with higher costs due to the utilization of several therapeutic strategies and more frequent appearance of disease progression/recurrence.Gynecologic Oncology 11/2010; 119(2):243-9. · 3.69 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The use of multiple medications is becoming more common, with a correspondingly increased risk of untoward effects and drug-related morbidity and mortality. We aimed at estimating the prevalence of prescription of relevant potentially interacting drugs and at evaluating possible predictors of potentially interacting drug exposure. We retrospectively analyzed data on prescriptions dispensed from January 2004 to August 2005 to individuals of two Italian regions with a population of almost 2.1 million individuals. We identified 27 pairs of potentially interacting drugs by examining clinical relevance, documentation, and volume of use in Italy. Subjects who received at least one prescription of both drugs were selected. Co-prescribing denotes "two prescriptions in the same day", and concomitant medication "the prescription of two drugs with overlapping coverage". A logistic regression analysis was conducted to examine the predictors of potential Drug-Drug Interaction (pDDIs). 957,553 subjects (45.3% of study population) were exposed to at least one of the drugs/classes of the 27 pairs. Overall, pDDIs occurred 2,465,819 times. The highest rates of concomitant prescription and of co-prescription were for ACE inhibitors+NSAIDs (6,253 and 4,621/100,000 plan participants). Considering concomitance, the male/female ratio was <1 in 17/27 pairs (from 0.31 for NSAIDs-ASA+SSRI to 0.74 for omeprazole+clopidogrel). The mean age was lowest for methotrexate pairs (+omeprazole, 59.9 years; +NSAIDs-ASA, 59.1 years) and highest for digoxin+verapamil (75.4 years). In 13/27 pairs, the mean ages were ≥70 years. On average, subjects involved in pDDIs received ≥10 drugs. The odds of exposure were more frequently higher for age ≥65 years, males, and those taking a large number of drugs. A substantial number of clinically important pDDIs were observed, particularly among warfarin users. Awareness of the most prevalent pDDIs could help practitioners in preventing concomitant use, resulting in a better quality of drug prescription and potentially avoiding unwanted side effects.PLoS ONE 11/2013; 8(10):e78827. · 3.53 Impact Factor
ASSET (Age/Sex Standardised Estimates of Treatment): A
Research Model to Improve the Governance of
Prescribing Funds in Italy
Giampiero Favato1,3*, Paolo Mariani2, Roger W. Mills1, Alessandro Capone3, Matteo Pelagatti2, Vasco Pieri3, Alberico Marcobelli4, Maria G.
Trotta5, Alberto Zucchi6, Alberico L. Catapano3,7
1School of Projects, Processes and Systems, Henley Management College, Henley-on-Thames, United Kindgom, 2Department of Statistical Science,
Bicocca University, Milan, Italy, 3Servizio di Epidemiologia e Farmacia Preventiva (SEFAP), Milan, Italy, 4Regional Health Authority (ASSR) Marche,
Ancona, Italy, 5Regional Health Authority (ASSR) Basilicata, Potenza, Italy, 6Local Health Authority (ASL) Milano 3, Monza, Italy, 7Department of
Pharmacological Sciences, University of Milan, Milan, Italy
Background. The primary objective of this study was to make the first step in the modelling of pharmaceutical demand in
Italy, by deriving a weighted capitation model to account for demographic differences among general practices. The
experimental model was called ASSET (Age/Sex Standardised Estimates of Treatment). Methods and Major Findings.
Individual prescription costs and demographic data referred to 3,175,691 Italian subjects and were collected directly from
three Regional Health Authorities over the 12-month period between October 2004 and September 2005. The mean annual
prescription cost per individual was similar for males (196.13 euro) and females (195.12 euro). After 65 years of age, the mean
prescribing costs for males were significantly higher than females. On average, costs for a 75-year-old subject would be 12
times the costs for a 25–34 year-old subject if male, 8 times if female. Subjects over 65 years of age (22% of total population)
accounted for 56% of total prescribing costs. The weightings explained approximately 90% of the evolution of total
prescribing costs, in spite of the pricing and reimbursement turbulences affecting Italy in the 2000–2005 period. The ASSET
weightings were able to explain only about 25% of the variation in prescribing costs among individuals. Conclusions. If
mainly idiosyncratic prescribing by general practitioners causes the unexplained variations, the introduction of capitation-
based budgets would gradually move practices with high prescribing costs towards the national average. It is also possible,
though, that the unexplained individual variation in prescribing costs is the result of differences in the clinical characteristics
or socio-economic conditions of practice populations. If this is the case, capitation-based budgets may lead to unfair
distribution of resources. The ASSET age/sex weightings should be used as a guide, not as the ultimate determinant, for an
equitable allocation of prescribing resources to regional authorities and general practices.
Citation: Favato G, Mariani P, Mills RW, Capone A, Pelagatti M, et al (2007) ASSET (Age/Sex Standardised Estimates of Treatment): A Research Model to
Improve the Governance of Prescribing Funds in Italy. PLoS ONE 2(7): e592. doi:10.1371/journal.pone.0000592
The Italian public funding of pharmaceutical prescribing is rapidly
evolving from a state centric model to one based upon the
equilibrium of central governance of demand and regional
funding. The state has exclusive power to define the basic
pharmaceutical coverage, which must be uniformly provided
across the country, while each Regional Health Authority (ASSR)
is responsible for funding the prescribing costs. Equity of access to
drug treatment on the basis of clinical need alone remains the
central principle of the national healthcare system, raising the issue
of an equitable distribution of resources in proportion to the
population needs. Since 2004, AIFA, the Agency for Italian Drug
Administration, is responsible for the governance of public
pharmaceutical prescribing. Besides its regulatory, pricing and
reimbursement functions, AIFA is responsible for maintaining the
level of public pharmaceutical spending below the threshold of
13% of total public healthcare costs. In case of overspending,
AIFA can apply generalised price reductions or modify the level of
national pharmaceutical coverage, by delisting entire classes of
drugs from the reimbursement list or by limiting the prescription
of reimbursed medicines to specified indications. Regional
Authorities (ASSR) cannot modify the level of pharmaceutical
coverage, but they are entitled to increase local taxes and to apply
a prescription fee in order to secure an adequate funding of
regional pharmaceutical demand .
Many ASSRs are considering introducing capitation based
prescribing budgets for their general practices. There are two
important factors driving this process: the first is cost containment.
It is assumed that budgets will encourage general practitioners to
examine their prescribing more critically, resulting in more cost
effective and appropriate prescribing. The second factor behind
the increasing interest in budgets is the belief that such budgets will
help to ensure that resources are allocated more fairly among
general practices. The implicit assumption is that, over a number
of years, practices will move towards the average and that
variation in prescribing costs between practices will be reduced
Academic Editor: Atle Fretheim, Norwegian Knowledge Centre for the Health
Received March 12, 2007; Accepted May 29, 2007; Published July 4, 2007
Copyright: ? 2007 Favato et al. This is an open-access article distributed under
the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the
original author and source are credited.
Funding: This study did not receive any grant or partial contribution from any
public or private organisation.
Competing Interests: The authors have declared that no competing interests
* To whom correspondence should be addressed. E-mail: gfavato@chicagogsb.
PLoS ONE | www.plosone.org1July 2007 | Issue 7 | e592
Weighted capitation based budgets seemed to offer the British
National Health Service (NHS) a solution to tackling the dual
problem of variations in prescribing costs and increasing drug costs
in general practice . Derived by the Prescribing Research Unit
(PRU) in 1993, the Age, Sex and Temporary Residents Originated
Prescribing Units (ASTRO-PUs) were designed to weight in-
dividual practice populations for age, sex and temporary residents.
The subsequent introduction of cost-based ASTRO-PUs to
allocate prescribing funds and to compare the costs of prescribing
between practices was widely criticised [4–6].
In particular, Smith argued that the formula did not reflect all
patients’ related variations in costs, random variations in clinical
needs, and differences in clinical practice .
Sheldon et al (1994) failed to find any convincing evidence that
factors of need other than age and sex were associated with
variations in healthcare utilization . The Specific Therapeutic
group Age/sex Related Prescribing Units (STAR-PUs), based on
British National Formulary (BNF) chapters, were introduced in
1995 as a way of accounting for differences in demography when
considering prescribing in different therapeutic areas. These
weights were reviewed and revised in 1997 . A study
commissioned by the NHS Executive examined the determinants
of NHS prescribing expenditures at practice level by relating costs
to population needs. The model was based on four variables:
permanent sickness, percentage of dependants in no carer
households, percentage of students, and percentage of births in
the practice list. Together with adjustments made for differences in
ASTRO (97)-PUs, the derived robust needs based capitation
formula was capable of explaining 62% of variations in prescribing
expenditures at practice level .
Understanding the determinants of demand for pharmaceuti-
cals is critical for a better assessment of the forces that increase
prescribing expenditures. Ageing and technological change play
a major role in this context with cohorts living longer that
consume increasing amounts of intensive, previously unavailable
treatments. More sophisticated econometric models recognised the
relevance of clinical determinants to the demand for prescribing,
such as morbidity and mortality standardised ratios, chronic illness
rates and physicians’ prescribing behaviour. Other socio-economic
factors, like patients’ disposable income, level of education and
access to healthcare, also influence the utilisation of pharmaceu-
tical treatments .
The primary objective of this study was to make the first step in
the modelling of pharmaceutical demand in Italy, by deriving
a weighted capitation model to account for demographic
differences among general practices. The experimental model
was called ASSET (Age/Sex Standardised Estimates of Treat-
ment). Most of the existing models of demographic predictors of
prescribing costs have been developed in England, where the
National Health Service (NHS) has been adopting capitation
based formulae, adjusted for age, sex, morbidity and socio-
economic factors for allocating prescribing budgets. Similarly to
NHS, the Italian Healthcare System is single payer based, but the
development of funding formulae has been delayed by the
availability of quality data at individual level. It is becoming
increasingly common for local (ASL) and regional (ASSR)
Healthcare Authorities to maintain a network of departmental
electronic databases, making it possible to integrate all relevant
information in the analysis of the trends of pharmaceutical
utilization. Most of socio-economic data are still available at
aggregate level, raising the need for an accurate age/sex
standardisation of prescribing costs among different cohorts. The
ASSET model provides a fundamental pre-requisite to a further
development of capitation based formulae at regional level.
Prescribing cost data have several characteristics that make
them a challenge to analyse. In this paper we discuss the
methodological and practical implications for policy makers,
healthcare administrators and general practitioners related to the
adoption of a weighted capitation formula, trying to answer the
following basic questions:
1.What is the formula to use to allocate prescription budgets to
general practices equitably?
How well does the formula explain the changes in total
prescribing costs over time?
How well does the formula explain the variability of
individual prescription costs in a single year?
If the formula were adopted, what would the main
implications be for policy makers and general practitioners?
The ASSET model (Age/Sex Standardised Estimates of
Treatment) model provides a research-based contribution to these
controversial issues in general practice.
Patient and cost data were obtained directly from participating
local (ASL) and regional (ASSR) healthcare authorities collecting
computerised prescription records for a two year period, from
January 2004 to December 2005. In particular, the demographic
database provided information on subjects’ date of birth, sex and
healthcare identification number, while prescription data (in-
cluding patient’s name and healthcare identification number, date
of issue, name of prescribing physician, brand name of the drug(s)
prescribed, generic name, ATC classification, cost and patient’s
co-payment) was extracted from the territorial pharmaceutical
All personal data (name and identification number) were
replaced by a univocal numerical code, making both databases
anonymous at source in strict compliance with the Italian Privacy
Law (Decree 196, 30/06/2003). The study design, observational
and retrospective in nature, did not require a previous informed
consent from the subjects included (Decree 196/03, art. 110).
The univocal numerical code, attributed to all subjects included
in the analysis, made it possible to retrospectively match
demographic patient’s data with individual prescription costs.
This study was based on data from a 12 month period, as
complete patient and cost data from all sources were available from
October 2004 to September 2005. No major change in
pharmaceutical demand or supply took place during the observa-
tion period. The total public expenditure for pharmaceuticals was
reported to be J12.4 billion in 2005 and J12.6 in 2004 .
The ASSET sample, collected from the ASL of Monza, the
ASSR Marche and the ASSR Basilicata, totalled 3,175,691
Table 1 shows the number of subjects and the percentage in
each age/sex group for the sample and for the 2005 population
estimates provided by the Italian National Institute of Statistics
Using the 2005 ISTAT figures as a base (see Table 1), the
difference between the sample and the expected distribution of
population could be calculated using a chi-square goodness of fit
test. The statistic was 17,064 with 15 degrees of freedom, which
was highly significant (p=0.000). The null hypothesis that the
ASSET sample and the Italian population had a similar age/sex
distribution could be rejected. The most significant differences
were in both tails of the distribution: the ASSET model had
a larger number of subjects aged ,14 and .75 compared to the
distribution of population. Although the differences in Table 1 are
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org2July 2007 | Issue 7 | e592
highly statistically significant, this is largely a function of the
number of patients in each cell of the table. The absolute
differences between the ASSET population and the Italian
population are small: similarly to STAR-PUs, since the ARTE
weightings were derived from the cost and number of items per
patient, rather than from the total costs and number of items, this
should not necessarily affect the quality of the measures .
Individual cost data referred to the pharmaceutical spending
over the 12-month period between October 2004 and September
2005. Pharmaceutical spending was defined as the total individual
cost for reimbursed drugs only (class A), dispensed by retail
pharmacies (not including hospital consumption), at actual prices
including co-payment. During the observation period, co-payment
was limited to a fixed prescription fee amounting to J1/J2
depending on the number of items per prescription. In 2005, total
co-payments amounted to 3.8% of total prescribing costs .
Excluded from the analysis were special drugs dispensed from
hospital pharmacies and out-of-pocket expenses for non-reim-
bursed drugs (class C).
The average prescribing costs by age group were calculated
simply by dividing the total pharmaceutical cost per age group by
the total number of subjects in the same age bracket. Note that the
number of subjects was the number registered in the demographic
database and not the number of persons receiving prescriptions.
Weights were obtained by dividing each average prescribing cost
by the total average cost (J195.6). Let i be the age/sex group,
ASSET weight for group i~
Sample total cost for group i
Sample population in age i
:Total population in sample
Total cost in sample
The ASSET weighting method was different from the one
adopted by the British STAR/ASTRO-PUs, where each average
cost by age group was divided by the average cost of the 0–4 age
group. The choice of total average cost as a weighting constant
was recommended by those Healthcare Administrators who
presently do not have access to computerised patient records.
While they cannot derive the average cost by individual age
groups, they can still calculate the total average prescribing cost
(total pharmaceutical expenditure divided by total assisted
population). The constant known, they can derive a pro-forma
age/sex weighted budget for each individual practice based on
prior year prescribing costs.
The cost data reported in Table 2 allow to easily calculating the
weights consistently to the STAR/ASTRO-PUs formula. A strong
caveat to any direct comparison between the two sets of weightings
cannot be overemphasized, as they reflect different reimbursement
and prescribing policies, drug prices, prescribing behaviours and
The ASSET model grouped all patients into16 age groups,
differently from the 18 weightings used by the STAR/ASTRO-
PUs. The 0–4 and 5–14 age groups were aggregated into a single
0–14 cluster, since in Italy children under 15 years are mandatory
seen by Paediatricians. Table 2 reports the breakdown of the 0–14
age cluster into 0–4 and 5–14 separate groups (in italic), but the
relative weightings were not utilised in the prescribing costs
The ASSET model
To derive a relationship between demography and prescribing
costs, individual cost data were collected for 3,175,691 subjects
living in three different regions of Italy (Lombardy, situated in the
north, Marche in the centre and Basilicata in the south). The
observation period was 12 months (Oct 2004–Sep 2005).
Table 1. Distribution of ASSET clusters compared to the distribution of Italian population (ISTAT, 2005)
ASSET sample sizeItalian population
Age group Male FemaleMale% Female% MaleFemale Male%Female%
,14 212,014198,0376.68% 6.24%4,242,0204,013,6927.26%6.87%
35–44 261,120254,4528.22%8.01%4,728,8444,678,965 8.09%8.00%
Total1,545,2171,630,474 48.66% 51.34%28,376,804 30,085,57148.54% 51.46%
Table 2. ASSET’s mean values by age group and standardised
weights for overall prescribing.
Mean cost (Euro)Standardised weights
Age groupMales FemalesMales Females
25–34 52.75 62.750.270.32
55–64300.88 277.401.54 1.42
.75652.75 481.20 3.34 2.46
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org3 July 2007 | Issue 7 | e592
The standardized weights given in Table 2 were obtained by
dividing the mean cost of each age group by the total mean cost
The distribution of both male and female subjects included in
the ASSET sample was non-normal, with a mode at 0 and a heavy
right tail. The mean annual prescribing cost per individual was
similar for males (J196.13) and females (J195.12), but the
distribution of medicine utilization by age showed significant
differences (above 10%).
Differently from what observed in England, in Italy prescribing
costs among young children (0–4 years) were lower than costs
among older children (5–14). This was not surprising: in the Italian
Healthcare system, expensive neo-natal treatments are directly
dispensed by hospital and local ASL and this specific drug
distribution was not captured by the ASSET cost data.
Compared to males’ mean costs of treatment, female drug
utilization was lower in the pubescent and teens age (first two
brackets), higher in the adult life (third and forth bracket), fairly
similar in the fifties (fifth bracket), to increase at a significant lower
rate than males in the senior years (from 55 to death). Taking into
account that neither contraceptives (not reimbursed) nor expensive
fertility drugs (delivered in hospital) were included in the
prescription costs analysed, the reasons for these time-lagged
discrepancies remained unexplained and they would be worth
After 65 years of age, the mean prescribing costs for males were
significantly higher than females. On average, a 75 year old
subject would cost 12 times a 25–34 years old one if male, and 8
times if female. Figure 1 shows that subjects over 65 years of age
(22% of total population) accounted for 56% of total prescribing
To evaluate the error embedded in the age/sex distribution of
the sample size, the ASSET weightings were used to estimate 2005
total prescribing costs. The cost estimate was calculated by adding
the costs by age/sex group obtained by multiplying the number of
Italian residents in each group by the relative weight and
subsequently by the actual mean pharmaceutical public spending
per resident published by AIFA (J211.5). The estimate of total
costs was then compared to the 2005 total prescribing costs,
reported by AIFA as J12.63 billion, including co-payment. The
ASSET age/sex model estimated 94% of total 2005 Italian
The main factor to cause the variance between the ASSET
weightings based estimate and the actual total Italian prescribing
cost was the cost of the non-assignable items, not included in the
calculation of weights (4.23% of total ASSET costs). Most of the
non-assignable items were prescribed to patients whose identity
was not included in the demographic databases, possibly either
because they were non-resident or because they just moved in
from a different healthcare district. The demographic databases
are aligned with census data at a given frequency; therefore a time
lag effect is always to be taken into consideration when measuring
the accuracy of cost data analysis for large cohorts. Less than 1%
of the total analysed prescription data were non-assignable due to
errors in compilation or missing data.
Testing the explanatory power of the ASSET model
The ASSET weightings were used to retrospectively explain the
evolution of total Italian prescribing costs of reimbursed drugs
(class A) between 2000 to 2005.This relatively short time frame
was chosen because age/sex weightings reflect clinical needs and
prescribing patterns that may change over a longer period of time,
due to the availability of improved diagnostic tools or innovative
therapeutic options .
Table 3 shows the 2000–2005 Italian resident population
weighted using the ASSET’s weights to highlight inequalities in the
age/sex distribution (ageing, adult immigration) that could have
had an impact on prescribing demand, measured as total public
spending on pharmaceuticals (including co-payment).
The explanatory power of the ASSET model was tested by
regressing the five year trend of total public pharmaceutical
expenditures and the ASSET-weighted population residuals. The
weighted Italian population was obtained by adding the number of
residents by each age/sex group multiplied by the relative ASSET
Figure 1. ASSET sample cumulative distribution of prescribing costs by age.
Table 3. 2000–2005 annual Italian resident population (ISTAT),
resident population weighted using the ASSET’s weights and
total pharmaceutical spending on reimbursed drugs,
including co-payment (AIFA).
Total cost (Mill
2005 58,462,375 55,018,91812,363
2004 57,888,245 54,129,40312,580
2003 57,321,070 53,259,66611,737
2001 56,960,69251,860,378 11,621
2000 56,923,52451,381,248 9,625
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org4 July 2007 | Issue 7 | e592
Weighted Italian population~
Residents in group i
Asset weight for group i
Weighted Italian population~
(residents by age=sex group|ASSET weight)
Due to the limited number of regression points, simply rejecting
the null hypothesis was not very informative, since the very low
power of the test would give a high probability of type II errors.
Although we used a 10% significance level, instead of the classical
5%, the significance tests must be read with caution, keeping in
mind that the degrees of freedom were too few to carry out
a meaningful inference.
The weighted population’s p-value (0.1028) was on the
boundary of the rejection region, but the F test (11.19) for the
overall regression was significant. The R-square was 0.88, but with
3 degrees of freedom was not surprising. The negative coefficient
of the weighted population was also expected, since it was certainly
collinear with the time trend, and its action in this model was to
correct for the age structure of the population.
Taking into account the incremental nature of prescribing costs,
we tested the impact of a time trend on the regression outcomes.
Adding a linear variable (years) to the regression did not change
the R-square (0.88) and the significance (F test: 11.18) of the
The interpretation of this regression model is that the total
public prescribing costs are growing faster than the ratio
underlying the ASSET model (constant spending for age/sex
groups) would predict.
Besides confirming the relevance of age and sex distribution as
predictors of prescribing demand, this outcome was particularly
interesting, taken into account the continuous effort operated by
AIFA to centrally maintain the level of pharmaceutical spending
below the threshold of 13% of total healthcare spending. A
number of actions were taken during those years to reduce
prescribing costs, such as:
i.De-listings of drugs from reimbursed status (Class A);
ii.Generalised price cuts;
iv.Direct distribution of innovative, expensive treatments;
The power of the ASSET model to explain individual cost
variability was determined by fitting a linear regression analysis
using the log transformed actual individual costs reported by
50,000 subjects randomly drawn from the sample, as the
dependent variable and the log transformed mean costs by age
group as the independent variable. Cost data are often
transformed to the log scale, which shortens the long right tail,
lessens heteroskedasticity, and decreases the influence of outliers
An equal probability sample of 50,000 patients was drawn from
the ASSET database using the randomization procedure of Oracle
Dynamic Sampling. The random sample included subjects
showing no prescription in the year (zero cost).
Due to the number of zero prescription costs in the sample, it
was assumed that the conditional distribution of the prescription
costs was a mixture of two distributions: a distribution that puts all
its probability mass in the point 0 and a log-normal distribution.
The maximisation of the likelihood could be split in two steps: first
a binary model to estimate the probability that a patient would
receive a prescription in the 12 month observation period (Y.0),
and then a log-normal regression model for the positive Y values
could be estimated.
The likelihood function of the model was:
For the binary model, the McFadden R-square was 0.13. All
model’s parameters were highly significant. Standard errors were
based on a robust Huber/White covariance matrix. The sign of
the regressor was positive as expected.
For the log-normal regression, the standard errors have been
calculated using White’s heteroskedasticity-consistent matrix, since
White’s test indicated possible heteroskedasticity. The R-square
(limited to persons with non-zero prescriptions) was 0.25. The
Jarque-Bera test for normality (as well as other common normality
tests) rejected the normality hypothesis, but this is not uncommon
with so many observations in sample.
The expected value of the prescription cost was given by:
Few extreme underestimations dominate the variance of the
prediction error [Y2E(Y|X)]: the maximum error was circa
J27,912, that was almost 20 times larger than the absolute value
of the minimum error (J1,483).
In order to check if alternative transformations of the two
variables could yield better results we estimated the double Box-
with Yi(l) and Xi(c) indicating Box-Cox transforms with param-
eters l and c , by Gaussian maximum likelihood. (Notice that since
we are supposing the conditional normality of a transform of our
data, we have to modify the likelihood according to the Jacobian of
this transformation. The correct form of the log-likelihood may be
found, for example, on page 500 of Green W. (1993) Econometric
Analysis, 5thedition, Prentice Hall.) Since we are supposing the
conditional normality of a transformation of our data, we have to
modify the likelihood according to the Jacobian form of this
transformation . The residual distribution and the R-square
(0.26) of this model are only slightly better then the previous one,
even though formal tests for l=0 and l=1 reject the null at any
usual significance level. By looking at the estimated Box-Cox
transform parameters, the logarithm seems a sensible choice for
the dependent, while a square root seems reasonable for the
regressor. Although the results are significant the difference
between the two estimated models seems not so relevant form
a practical point of view.
Beyond the technicalities of the analysis, the important finding
was that the ASSET model could not explain large variations in
individual prescribing costs. This was not a surprising outcome: in
regression analyses of healthcare utilization data, the R-square
values were usually on the order of ,20%. Newhouse used
theoretical and empirical arguments to estimate that the maximum
possible R-square was about 48% for outpatient costs .
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org5 July 2007 | Issue 7 | e592
As discussed earlier, to increase the explanatory power of
capitation based models, additional determinants of prescribing
demand should have been considered, such as morbidity and
mortality ratios, chronic illness rates, deprivation and access to
healthcare, together with other relevant socioeconomic determi-
nants, like disposable income and level of education. The ASSET
model was a first step in developing a more rational approach of
allocating prescribing funds in Italy.
Implications for Italian healthcare policy makers
Policy makers not only need to know the determinants of public
prescribing expenditures, but they should also have the possibility
to estimate the impact of those trends having a significant impact
on pharmaceutical demand.
The ageing of population is a known fact. According to ISTAT
data, in the last twenty years, life expectancy at birth increased by
6 years for males (76.9 years) and by 5 years for females
(82.9 years). The ASSET model confirms the strong, quasi-
exponential relationship between age and pharmaceutical utiliza-
tion, allowing policy makers to quantify the impact of ageing
population in terms of resources needed to satisfy the incremental
As an example, the intermediate scenario of the latest
population projections foresees in 2026 a marginal decrease in
the total number of Italian residents (57.5 million), down by 1.6%
compared to the current 58.5 million inhabitants . All else
equal (prices, therapeutic alternatives, and public coverage of
prescription costs), we could assume that prescription costs should
remain relatively stable over the next two decades. The ASSET
model helps policy makers and demographic statisticians to
actually demonstrate the opposite.
Multiplying the ASSET weights by the expected number of
residents, we obtain an age/sex-standardised population that
reflects the relative cost of pharmaceutical utilization. A 0–14 year
old male, on average, accounts for just one fifth of the mean
annual prescribing cost, while a 75 year old male uses 3.3 times as
many medicines as the average. Population data standardised with
the ASSET’s weights represent a close proxy of pharmaceutical
spending. In table 4 we derive the weighted Italian population
estimated for 2026 using the ASSET weights. The weighted
population is expected to grow from 55 million in 2005 (see
Table 3) to 65.8 million in 2026. All else equal, the pharmaceutical
spending in 2026 is likely to be almost 20% higher than in 2005 as
a result of the ageing population.
The ageing process shows wide regional variability. Regions
with the highest percentage of residents over 65 years old are
located in the North and Centre of Italy: Liguria (26.2%), Umbria
(23.1%), Toscana (22.9%), Friuli and Piedmont (21.8%). The
Southern regions show the lowest percentage of elderly residents:
Campania (14.7%), Puglia (16.5%) and Sardinia (16.6%).
Policy makers must allocate adequate resources to regions to
fund prescribing costs based on clinical needs rather than
population density. A simple capitation formula would ignore
differences in demographic distribution among regions, inevitably
under-funding those areas with the highest concentration of
elderly population. For example, consider two Italian regions,
Liguria (North East) and Sardinia (island), that have a similar
population of 1,6 million and 1,7 million resident respectively, but
a ten percent points difference in the elderly population (22.9% vs.
16.6% residents over 65 years old).
A straight capitation formula would allocate to Liguria
a prescription budget 3.6% lower than the one allocated to
Sardinia. Comparing the number of residents weighted by age and
sex (1.8 million in Liguria vs. 1.5 million in Sardinia), we realise
that Liguria actually needs 18% more prescribing funds than
Sardinia to cover the therapeutic needs of its older population
Table 4. 2026 Italian resident population projected by ISTAT (intermediate scenario) weighted using the ASSET’s weights.
Age groupsEstimated Italian population in 2026ASSET weights2026 weighted population
Male FemaleMales FemalesMales Females
15–242,909,978 2,759,2860.230.21 668,435 577,535
25–34 3,083,1462,958,0390.270.32 831,462948,853
45–54 4,237,5664,133,935 0.75 0.763,167,1313,162,039
55–644,545,376 4,587,1541.541.426,991,435 6,505,063
.753,048,694 4,805,8013.342.46 10,173,43211,822,110
Total 28,037,46629,484,162 32,534,82133,230,996
Table 5. Comparison of 2005 population of two Italian regions
(Liguria and Sardinia) weighted using the ASSET’s weights.
Age groupsLiguria’s weighted residentsSardinia’s weighted residents
Male FemaleMale Female
,1418,916 15,47623,618 19,063
15–24 14,219 12,28922,924 19,761
25–34 26,811 30,92335,436 40,688
35–44 51,603 57,16355,30162,074
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org6 July 2007 | Issue 7 | e592
The ASSET model is a useful tool to support the process of long
term healthcare policy planning as well as the equitable allocation
of annual prescribing resources to regional authorities.
Implications for regional healthcare administrators
Regional healthcare authorities (ASSR) could use a similar
mechanism to equitably allocate prescribing cost guidelines to
general practices on the basis of population need.
Should two practices with a similar number of patients each
receive the same level of prescribing funds? Not necessarily. The
demographic differences in general practices ought to be recognised
by the formula used to allocate the pharmaceutical budget.
Let’s compare two general practices, A&B. Both practices have
the maximum number of patients (1,500) allowed by the Italian
Healthcare System and the proportion of male and female patients is
approximately the same for both practices (47% male vs. 53%
female). The age distribution, though, differs significantly between
the two practices: 935 patients (62% of total) in practice A are older
than 65, compared to 642 patients (43% of total) in practice B.
Table 6 shows that while a simple per capita distribution of resources
would allocate to both practices an equal prescribing budget , the
ASSET model reflects both the size of the practice list (1,500 patients
each) and its age and sex structure in the budget allocation , granting
to practice A (3,020 weighted patients) a prescribing budget 13%
higher than practice B (2,621 weighted patients).
Implications for general practitioners
If health authorities are considering the introduction of capitation
based budgets then general practitioners will need to prepare for
this. They should familiarise themselves with the basic methods of
cost analysis in order to understand the factors that can increase
the demand for medicines and to be able to discuss the wide
variations in prescription costs among patients.
Demographic adjusted healthcare cost models, such as ASSET,
tend to lose their explanatory power when the subset of population
examined gets smaller. As discussed in the previous section, while
the ASSET model is able to explain 60% of the five-year
pharmaceutical cost trend of Italy, a country of 58 million
inhabitants, its ability to explain individual utilization differences
goes down to approximately 30% for a sample of 3.1 million
subjects. When the unit of analysis is a single practice,
idiosyncratic prescribing causes may overshadow differences in
clinical characteristics of practice population, such as the incidence
of diabetes, asthma or ischaemic heart disease.
A weighted capitation base formula would then classify practices
only as low cost or high cost prescribers, telling nothing about the
quality of prescribing, an essential determinant of demand. This
information can only come from a detailed analysis of practice’s
prescribing data combined with information directly collected
from each practice.
The implementation of the ASSET capitation formula would
provide an effective benchmark to compare prescribing costs
standardised by age and sex differences in the practice list, but is
just a starting point in the process of optimization of prescribing
resources It should help practitioners to reflect upon specific
determinants of demand for medicines in their practice, such as
the transfer of care from hospitals to general practice or a high
prevalence of chronic diseases, to identify areas in which costs
could be saved through a more rational prescribing.
The ASSET age/sex standardisation model, therefore, proved to
be a useful, but not an exhaustive tool to equitably align the
distribution of resources among regions according to their relative
The ASSET weightings were able to explain only about 25% of
the variation in prescribing costs among individuals: the causes of
the remaining 75% variation in prescribing costs remained
unknown. The magnitude of individual variance was extremely
significant: the individual costs value in the ASSET sample ranged
between 0 and .40,000 euros. The ASSET sample included the
registered persons who did not receive any prescription in the
same time period: 808,464 subjects (26% of the total sample) did
not receive a prescription, of whom 488,120 males (32% of total
males) and 320,344 females (20% of total females).
From a different perspective, the ranking by total pharmaceutical
annualcostof the 50,000individualsincluded inthe randomlydrawn
highest spending subjects was associated with 51.4% of total
pharmaceutical spending. The derivation of a robust model capable
of identifying the drivers of individual variances should be the
objective of further research. If mainly idiosyncratic prescribing by
general practitioners causes the unexplained variations, the in-
troduction of capitation-based budgets would gradually move
practices with high prescribing costs towards the national average.
It is also possible, though,that the unexplained individual variation in
prescribing costs is the result of differences in the clinical
characteristics of practice populations or because some general
practices are better at early diagnoses, treatment and compliance of
based budgets may lead to unfair distribution of resources .
The ASSET age/sex weightings should be used as a guide,
not as the ultimate determinant, for an equitable allocation of
prescribing resources to regional authorities and general practices.
Conceived and designed the experiments: GF VP. Analyzed the data: GF
PM MP VP AC. Contributed reagents/materials/analysis tools: PM MP
RM AC VP AM MT AZ. Wrote the paper: GF. Other: Contributed to the
revision of the final submitted paper: RM AC MP VP AM MT AZ AC
Table 6. Hypothetical prescribing budget allocation using the
ASSET weights to two practices (A&B) with equal number of
patients, similar male/female ratio, but significantly different
Age groups Actual population Weighted population
Practice A Practice BPractice A Practice B
Male FemaleMale Female Male Female Male Female
15–24 1210 555032 1311
25–34 2226486068 1316
35–44 434863 7318 222630
45–5465687512848 5256 96
55–64 134 137144162 206195221249
Total 7047967087921,5531,467 1,281 1,340
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org7 July 2007 | Issue 7 | e592
1. Giarda DP, Petretto A, Pisauro G, Lorenzini S, Vignocchi C (2005) Elementi
per una politica di governo della spesa pubblica. Paper presented at the
conference ‘‘Oltre il Declino’’, Fondazione Rodolfo Debenedetti, Rome,
February 3. Available online at www.frdb.org.
2. Majeed A, Head S (1998) Capitation based prescribing budgets will not work.
BMJ 316: 748–750.
3. Roberts SJ, Harris CM (1993) Age, sex, temporary resident originated
prescribing units (ASTRO-PUs): new weightings for analysing prescribing of
general practices in England. BMJ 307: 485–8.
4. Coulter A (1994) Capitation funding may over fund practices in better off areas.
BMJ 308: 207–208.
5. Baker Y (1998) Use of capitation formulas for primary care groups could result
in chaos. BMJ 317: 210.
6. Carr-Hill R, Roberts D (1999) Population figures for capitation formulas need to
be designed differently. BMJ 318: 1145.
7. Smith P, Sheldon TA, Carr-Hill RA, Martin S, Peacock Hardman G (1994)
Allocating resources to health authorities: results and policy implications of small
area analysis of use of inpatient services. BMJ 309: 1050–4.
8. Sheldon TA, Smith P, Borwitz M, Martin S, Carr-Hill R (1994) Attempt at
deriving a formula for setting general practitioner fund holding budgets. BMJ
9. Lloyd DC, Harris CM, Roberts DJ (1995) Specific therapeutic age-sex related
prescribing units (STAR-PUs): weightings for analysing general practices’
prescribing in England. BMJ 311 (7011): 991–4.
10. Rice N, Dixon P, Lloyd DCEF, Roberts D (2000) Derivation of a need based
capitation formula for allocating prescribing budgets to health authorities and
primary care groups in England. BMJ 320: 284–88.
11. Majeed A, Malcolm L (1999) Unified budgets for primary care groups. BMJ.
12. Agenzia Italiana del Farmaco AIFA (2005) L’uso dei farmaci in Italia.
OSMED reports 2000–2005. Available online at: http://www.agenziafarma-
TORIALE§ion_code = AIFA_PUB_RAP_OSMED&cache_session =
13. Istituto Nazionale di Statistica (ISTAT) Demographic statistics analysis on the
Italian population 2004/2005. Available online at: http://demo.istat.it/
14. NHS. ASTRO-PUs and Chapter STAR-PUs: 2001 revision paper. Available
online at: http://www.ic.nhs.uk/psu/measure/presmeas/copy_of_atropus.
15. Smith PC (1999) Setting budgets for general practice in the new NHS. BMJ 318:
16. Diehr P, Yana D, Ash A, Hornbook M, Lin DY (1999) Methods for analyzing
healthcare utilization and costs. Annual Review of Public Health 20: 125–14.
17. Green W (1993) Econometric analysis. Prentice Hall, 5thedition. pp 500.
18. Newhouse JP, Manning WG, Keeler EB, Sloss EM (1989) Adjusting capitation
rates using objective health measures and prior utilization. Healthcare Finance
Review 10: 41–54.
19. Istituto Nazionale di Statistica (ISTAT) (2051) Demographic projections of the
Italian population by age, sex and region: 2001/2051. Available online at:
20. Whynes DK, Baines DL, Tolle KH (1996) Explaining variations in general
practice prescribing costs per ASTRO-PU (age, sex and temporary resident
originated prescribing unit), BMJ 312: 488–489.
Capitation of Prescribing Cost
PLoS ONE | www.plosone.org8 July 2007 | Issue 7 | e592