Disentangling spillover e¤ects of antibiotic consumption:
Published in Applied Economics (2013), 45, 8: 1041-1054,
Literature on socioeconomic determinants of antibiotic consumption in the
community is limited to few countries using cross-sectional data. This paperanalyses regional variations in outpatient antibiotics in Italy using a balancedpanel dataset covering the period 2000-2008. We specify an econometric modelin which antibiotic consumption depends upon demographic and socioeconomiccharacteristics of the population, the supply of health care services in the commu-nity, and antibiotic copayments. The model is estimated by means of Ordinaryleast squares techniques with …xed e¤ects (FE). The implications of consumptionexternalities across geographical areas are investigated by means of spatial-lagand spatial-error models (SLFE and SEFE). We …nd signi…cant and positive in-come elasticity and negative e¤ects of copayments. Antibiotic use is also a¤ectedby the age structure of the population and the supply of community health care.
Finally, we …nd evidence of spatial dependency in the use of antibiotics acrossregions. This suggests that regional policies (e.g. public campaigns) aimed atincreasing e¢ ciency in antibiotic consumption and controlling bacterial resis-tance may be in‡uenced by policy makers in neighbouring regions. There willbe scope for a strategic and coordinated view of regional policies towards theuse of antibiotics.
antibiotic consumption; socioeconomic inequalities; spatial de-
JEL classi…cation: C21; C23; I11; I18; R00
Department of Economics, University of Lugano, 6900 Lugano, Switzerland
yDepartment of Economics and Technology Management, University of Bergamo, 24044 Dalmine
(BG), Italy; Department of Economics, University of Lugano, 6900 Lugano, Switzerland. Corre-sponding author. E-mail: email@example.com. We are grateful to Massimo Filippini whosecomments and suggestions generated signi…cant improvements to the paper. Comments and sugges-tions from participants to the 11th National Health Economics Conference of the Portuguese HealthEconomics Association, Porto, 8-10 October 2009, and the 10th National Health Economics Con-ference of the Italian Health Economics Association, Bergamo, 29-30 October 2009, are gratefullyacknowledged. The usual caveat applies.
The increasing use of antibiotics and the consequent harmful e¤ects of bacterial re-sistance represent a growing problem in many countries. Evidence suggests thatbacterial resistance grows with antibiotic use (Monroe and Polk, 2000; Mera et al.,2006). Although the e¤ects of intervention policies on resistance to antimicrobialdrugs cannot be assessed accurately at present, public interventions may be e¤ectivein controlling antibiotic consumption (Huttner et al., 2010). Consequently, the in-vestigation of socioeconomic inequalities in the use of antibiotics across geographicareas is an important approach in understanding causes of consumption and buildinge¤ective intervention policies.
Literature shows that outpatient antibiotic consumption, measured by the num-
ber of de…ned daily doses per 1000 inhabitants (DID), is highly heterogeneous acrossEuropean countries (Elseviers et al., 2007). For instance, the consumption of an-timicrobials in France is almost three times the consumption in the Netherlands.
Generally, southern European countries exhibit higher levels of consumption as com-pared to northern European countries.
Outpatient antibiotic consumption is also highly heterogeneous across geographic
areas within a country (Kern et al., 2006). To our knowledge, the study of socioe-conomic determinants of consumption has been applied to few countries (Matuz etal., 2005; Filippini et al., 2006; Nitzan et al., 2010). Since previous studies are basedon cross-sectional data, unobserved heterogeneity may stem from omitted commonvariables that a¤ect geographic areas di¤erently. These latent common factors mayinduce cross-section dependence and lead to inconsistent estimation coe¢ cients in re-gressions if hidden aspects are correlated with the explanatory variables (see Cameronand Trivedi, 2005).
Spatial econometric approaches to antibiotic consumption are lacking.
graphic areas are usually treated as isolated entities, ignoring the fact that antibioticconsumption is plausibly a¤ected by consumption in neighbouring regions. Spatialinteractions in panel data are considered, for instance, by Revelli (2001) to investi-gate variations in tax rates across English districts, and by Costa-Font and Moscone(2008) to study health expenditure across Spanish regions. As for antibiotics, spatialaspects are partially addressed in recent studies by Filippini et al. (2009a, 2009b)using cross-sectional data. Spatial dependency plays an important role in the useof antibiotics for two main reasons. First, antibiotics are used to cure infectionswhich may spread to other individuals in the community. Second, antimicrobial re-sistance partially generated by the intensive use of antibiotics may reduce antibiotice¤ectiveness for other individuals in the community. It follows that regional policies(e.g. public campaigns) aimed at increasing e¢ ciency in antibiotic consumption andcontrolling bacterial resistance could blunt the impact of policies in neighbouringregions through the generation of local spillovers. This may suggest that the lack ofcoordination of regional policies towards the use of antibiotics leads to ine¢ ciency.
The purpose of this study is to investigate socioeconomic factors a¤ecting re-
gional variations in outpatient antibiotic use in Italy over a relatively large period of
time (2000-2008), by means of panel data analysis which takes the external e¤ects ofconsumption into account. The use of panel data makes it possible to specify …xedregional e¤ects in order to take the unobserved heterogeneity into account. A noveltyof our work is that it focuses on a country, Italy, of which socioeconomic determinantsof antibiotic use in outpatients have not been investigated so far. The relevance ofthis focus also originates from the peculiar organization of the Italian health caresystem, which is based upon a National Health Service where the provision of healthcare is substantially devolved to regional health authorities. A question arises asto whether the e¤ects of main socioeconomic determinants of consumption are sim-ilar in countries with di¤erent health care organizations, since previous studies areconducted in health care systems based upon health insurance plans.
According to a recent report of the European Commission (2010), Italy is the
most consuming country of antibiotics in Europe with relatively poor levels of pub-lic awareness of antibiotic e¢ cacy. This suggests that the investigation of factorsa¤ecting the use of antibiotics may raise more concern compared to other countries.
Similarly to previous studies, our model hypothesises that regional consumption ofoutpatient antibiotics in Italy depends on antibiotic price (copayments), populationage structure and income, the supply of health care services in the community, andthe health status of the population. The model is estimated by means of Ordinaryleast squares with …xed e¤ects (FE). Consumption externalities between regions areinvestigated by means of spatial-lag and spatial-error models with …xed e¤ects (SLFEand SEFE). Spatial lags may re‡ect interactions between regions whereas spatial er-rors occur because regions have unobserved factors in common.
The remaining of the paper is organized as follows. Section 2 provides an overview
of the literature on socioeconomic determinants of antimicrobial use across geo-graphic areas. In Section 3 we summarise the main features of the Italian market forprimary care and antibiotic use in outpatients. The speci…cation of the economet-ric model and the estimation approaches are presented in Section 4 and Section 5,respectively. Section 6 discusses the results and Section 7 concludes.
A review on socioeconomic determinants of antibioticconsumption
The literature on socioeconomic determinants of outpatient antibiotic consumptionis limited to few empirical studies, although there are more general studies on thedeterminants of pharmaceuticals use (e.g. Costa-Font et al., 2007). We present …vestudies which investigate the impact of socioeconomic factors within countries.1 Themain features of these studies - geographical setting, type of data, methodologicalapproach, determinants cosidered and results - are summarised in Table 1.
1 Regarding cross-country studies, we refer the reader to the recent analysis by Masiero et al.
Nitzan et al. (2010) analyse the use of antibiotics in outpatients in 8 districts
of Israel. The authors investigate consumption measured in de…ned daily doses per1000 inhabitants for di¤erent age groups and for di¤erent groups of antibiotics. Thestudy shows a decline in antibiotic use in all districts between 2003 and 2005 andlarge variation between districts. Results reveal that during the 3 years of the studythe highest antibiotic consumption rates are observed for the youngest age groups(0-4, 5-18, and 19-44). Antibiotic consumption among individuals aged 65 or aboveis by far the lowest in all the age groups, probably due to higher hospitalization rates.
Also, there is a signi…cant association between a higher prevalence of diseases, suchas diabetes mellitus, and higher antibiotic consumption. Conversely, higher ratesof hospitalization seem to be correlated with lower levels of antibiotic consumption.
Finally, the authors …nd large variability between the districts in the use of speci…cantibiotic groups. The use of penicillins in high consumption districts, for instance,is 2.8 times the use in low consumption districts. The magnitude of di¤erences raisesto 3.9 in the use of …rst-generation cephalosporines.
To investigate socioeconomic determinants of outpatient antibiotic use in Switzer-
land, Filippini et al. (2006) use regional consumption data and regress them againsta set of variables suggested by the literature as plausible causal factors of the demandfor drugs. The dataset includes quarterly data for 3 years (2002-2004) detailed atcantonal level (26 cantons). Findings show that Switzerland uses relatively low vol-umes of antibiotics in ambulatory care compared to other European countries, butlarge di¤erences are observed across cantons. The authors specify an ad-hoc demandfunction for the cantonal per capita outpatient antibiotic use which depends upon thehealth status of individuals, income, antibiotic price, age, education, density of physi-cians and cultural aspects summarised by linguistic groups and borders with othercountries. Since individuals health status and antibiotic price can be endogenous,the authors consider the inclusion of lagged values in the model and apply an instru-mental variable approach. The per capita income, antibiotic price, the proportionof foreign residents, the density of medical practices, and cultural and educationaldi¤erences are signi…cant determinants of consumption. Among these results, it isworth noting that income has a positive impact on consumption, and antibiotic pricehas a negative and signi…cant e¤ect, as expected. Conversely, physician density isassociated with higher levels of antibiotic use.
Kern et al. (2006) carry out an exploratory analysis on antibiotic prescriptions
in Germany for the year 2003. They investigate variations in outpatient antibioticsbetween 23 areas in 16 states both for overall use and for the use of speci…c classes ofantibiotics. Relatively low antibiotic consumption is observed in eastern and southernregions. Basic penicillins are the most frequently prescribed drugs with large regionalvariation. Regional patterns of use are similar for children and adults, although lowerlevels of consumption for children are observed in the south. The study does not …ndan association between overall antimicrobial consumption and population density,the percentage of elderly people, income, unemployment, gross domestic productand aspects of local health care supply, but the analysis lacks of a sound econometricapproach.
Filippini et al. (2009a) investigate ine¢ ciencies in the use of outpatient antibi-
otics across small areas. The authors carry out econometric estimations using atwo-stage least squares procedure on quarterly data of antibiotic use (in DID) inSwiss outpatients available for 240 small areas in 2002. A model is proposed inwhich antibiotic use varies according to the socioeconomic and demographic charac-teristics of the population, the incidence of infections, the local supply of health careand antibiotic price. The results suggest a positive relationship between antibioticconsumption and income, the proportion of children between 0 and 14 years of age,the percentage of foreigners in the total population, the incidence of infections, anddensity of pharmacies and physicians. On the other hand, antibiotic price and theproportion of individuals over 74 years of age show a negative and signi…cant impacton antibiotic use. Some seasonal e¤ects are found, which suggest that the per capitaoutpatient antibiotic use is lower in spring and summer periods. Finally, the authorsconsider the e¤ects of spatial dependency in antibiotic consumption across the areasby means of spatial lags included in their model. The negative impact of antibioticuse in neighbouring areas suggests that the use of antibiotics in one area may reducethe spread of infections in neighbouring areas.
Finally, Matuz et al. (2005) investigate regional variations in antibiotic con-
sumption in ambulatory care in Hungary. The sample is composed of 19 regions(counties) for the years 1996-2003. The authors …nd that antibiotic consumptionwas 21.1 DID in 1998, close to the European average, but decreases from 2002. Thestudy shows large and stable interregional variations in consumption. The ranking ofregions according to total antimicrobial consumption is basically the same during thewhole period. The authors test associations between total antibiotic consumptionand possible determinants of use by means of the two-tailed Spearman coe¢ cient fornon-parametric correlations. They do not …nd any signi…cant relationship betweenantibiotic consumption and the average monthly net income nor with the demo-graphic structure of the population. Conversely, a signi…cant association with totalantibiotic consumption is observed with the proportion of individuals receiving freeaccess to selected medicines from the public health system without quantity limitand the proportion of individuals regularly receiving social assistance.
Outpatient antibiotic consumption within the ItalianNHS
The Italian health care system is based upon a national health service (SSN) mainly…nanced by general taxation and characterised by universal access to health care forthe entire population and asymmetric decentralization of health care provision to the20 regions. Reforms over the 90’s gave administrative and …nancial responsibility inthe provision of health care to the regions. The central government retains limitedsupervisory control and continues to hold overall responsibility for the SSN to assureaccess and equal levels of health services across the country. The regions organiseservices that are designed to meet the needs of their local populations, de…ne ways
Figure 1: Outpatient antibiotic use in Europe by country (2000-2007). Data source:European Surveillance of Antimicrobial Consumption (ESAC).
to allocate …nancial resources to the local health authorities (LHAs) within theirterritories, monitor health care services and activities provided by LHAs, and assesstheir performance.
Outpatient care in Italy is provided by general practitioners (GPs), paediatricians
and specialists. General practitioners and paediatricians deliver primary care andpreventive medicine and are mainly paid on a capitation basis. Generally, patientsdo not pay for visits to GPs. They also enjoy considerable freedom of choice ofproviders since they are only obliged to use providers in the province in which theyreside and they must have a doctor’s prescription for most forms of care. Patientscan always change their GP within the province of residence (Atella et al., 2003).
Specialists delivering outpatient care are paid on a fee-for-service basis. Patients payonly a small fraction of the full cost of a consultation if they are referred by theirGP.
Medicines in outpatient care are classi…ed in two categories. The …rst category
(class A) includes essential medicines and medicines for serious and chronic diseaseswhich require a doctor’s prescription. These drugs are fully reimbursed by the SSN,although patients bear small copayments in some regions. Antibiotics are generallyincluded in this category and the copayment (ticket) includes both a cost-sharing
Figure 2: Outpatient antibiotic use in Italy by region (2000-2008). Data source:Italian National Observatory on Drugs Utilisation (OSMED).
scheme and a reference pricing one.2 According to this, patients are required tocontribute to the cost of antibiotics either by a …xed amount per prescription orby a proportional-to-…nal price amount, or by paying the di¤erence between the…nal price and the reference price.3 Generally, people with chronic or rare diseases,disabled people, pregnant women and low income people bene…t from exemptions.
The reference price is set for drugs that contain the same active ingredient, identicalpharmaceutical dosage and package size. The second category of drugs - the so-called“class C” - includes medicines for minor diseases and ailments, medicines the use ofwhich is discouraged and those not requiring a medical prescription. Pharmaceuticalproducts included in this category are not reimbursed by the SSN.
As concerning the use of outpatient antibiotics, Italy is a relatively high-con-
suming country. Using available data from the European Surveillance of Antimi-crobials Consumption (ESAC) between 2000 and 2007 we can rank Italy amongEuropean countries according to the number of de…ned daily doses per 1000 inhab-itants consumed (Figure 1). Italy ranks among the most consuming countries, just
2 In Italy only 1% of antibiotic courses are obtained from a pharmacy without a prescription.
3 See Fiorio and Siciliani (2010) for details.
Figure 3: Outpatient antibiotic use in Italy by antibiotic categories (2000-2008).
Data source: Italian National Observatory on Drugs Utilisation (OSMED).
below France, Greece, Luxemburg and Slovakia. In contrast, the Netherlands, Aus-tria and Denmark are among the least consuming countries. A recent survey by theEuropean Commission (2010) indicates that Italy had the highest rate of antibioticuse across Europe between December 2008 and November 2009. Moreover, the ma-jority of the population (51%) thinks that antibiotics are e¤ective against commoninfections, such as colds or ‡u, which are not cured by antibiotics. This evidenceraises concern about the e¢ cient use of antibiotics in the country.
The mean level of consumption of outpatient antibiotics in Italy between 2000
and 2008 was 23.53 DID, with a peak in 2008 (25.01 DID) and a minimum in 2000(22.36 DID).4 Figure 2 shows that antimicrobial use has been roughly stable over thisperiod but a remarkable degree of heterogeneity in consumption is observed acrossthe regions. Generally, regions in central Italy use more antibiotics per capita (24.94DID) than regions in the north (18.43 DID) and less than southern regions and theislands (28.74 DID).
The largest mean share of consumption during the time period considered is rep-
resented by the combinations of penicillin with beta-lactamase inhibitor (25.76%),
4 Data are collected by the Italian agency for drugs (AIFA).
followed by the macrolides (21.57%). Other categories of antimicrobials predomi-nately used in the community are the broad-spectrum penicillins (20.59%) and thequinolones (13.64%). The structure of consumption has slightly changed over the 8-year period (Figure 3). The share of combination of penicillin increased from 18.88%in 2000 to 31.10% in 2008. This increase was partially compensated by a decrease inbroad-spectrum penicillins, from 22.45% in 2000 to 17.20% in 2008. In this respect,it is worth noting that changes in the use of antibiotic classes are associated to sub-sequent changes in bacterial resistance. It has been observed that current penicillinresistance depends on the cumulative consumption in the previous two years (Al-bricht et al., 2004). The substitution of broad-spectrum penicillins with combinationof penicillin with beta-lactamase inhibitors is at least partially explained by the re-duced e¤ectiveness of the former category due to increasing bacterial resistance andthe availability of a more dynamic subclass of penicillins (Ferech, et al., 2006).
We propose a reduced-form consumption function for outpatient antibiotics,5 wherethe variability in antibiotic use among Italian regions depends on socioeconomiccharacteristics of the population (age structure and income), the supply of healthcare, individuals health status, and the price (copayment) of antibiotics. Antibioticconsumption in di¤erent periods is assumed to be fully separable (Bretteville-Jensen,2006). This means that current consumption a¤ects consumer’s utility in the currentperiod only, without consumer’s preferences interactions across time periods.
DIDit = f (Yit; Pit; DP Hit; P OP 1it; P OP 2it; IN Fit),
where the subscript i denotes the region and t the time period. DIDit is antibioticconsumption per capita; P OP 1it is the proportion of the population below 25 yearsand P OP 2it indicates the proportion of the population older than 64. DPHit is thedensity of general practices and IN Fit is the mortality rate from infectious diseases,which is a proxy for individuals health status. Finally, the model includes income(Yit) and copayment for pharmaceuticals (Pit).6
For the estimation of equation (1) we use a “hybrid” log-log functional form.7
The log transformation is applied to income and copayment only since the other
5 Within the Italian NHS, outpatient antibiotic consumption originates from prescriptions by
general practitioners. We hypothesize that the demand for antibiotics for patients registered in theGP’s list is de…ned by the representative general practitioner in the region by taking patients’needsand preferences into account.
6 There is evidence that income is strongly associated with education and is preferred as a deter-
minant of inequalities in the use of health care services (Habitch et al., 2009).
7 A linear functional form has also been considered. In this case, the results are less satisfactory
in terms of goodness of …t and the signi…cance of the coe¢ cients.
covariates are de…ned as percentage ratios. Equation (1) can then be written as:
5 capture the e¤ect of the young and the old population,
respectively, as compared to the reference population group aged between 25 and 65.
it is an error component with standard normality assumptions.
Our dataset is a balanced panel which includes data for 9 years, from 2000 to 2008,
for 20 Italian regions. Summary statistics are reported in Table 2. Data on regionaloutpatient antibiotic consumption are collected from annual reports prepared bythe Italian National Observatory on Drugs Utilization (Osmed). The per capitaconsumption is measured by the number of de…ned daily doses (DDD) per 1000inhabitants per day (DID). A DDD represents the standard dose necessary for oneday of drug treatment in adults and is de…ned by an independent scienti…c committeeanswering to the WHO Collaborating Center for Drug Statistics Methodology. TheDID measure can be interpreted as the number of persons (out of 1000) who aretaking antibiotics on a given day (Monnet et al., 2004). The dataset comprises salesof antimicrobial drugs - group J of the Anatomical Therapeutic Chemical (ATC)classi…cation - included in class A by the Italian NHS (see Section 3 above). Thesedrugs require a doctor’s prescription and are supplied virtually free of charge, againstsmall patient copayments.
Information on the demographic structure of the population, per capita income,
density of general practices and mortality from infectious diseases are obtained fromthe Italian National Institute of Statistics (Istat). Copayments are obtained fromannual reports on pharmaceutical consumption and expenditure prepared by Osmed.
Regional copayments vary from 0 to 4 Euros. We rescale the variable from 1 to 5in order to avoid negative values in the log transformation.8
infectious diseases in 2004 and 2005 are not available. Since data are substantiallystable before and after this period, we linearly interpolate mortality data from 2003and 2006 to assign values to the years 2004 and 2005. We are aware that thisapproximation could create some bias in the estimation. We then carefully discussthe estimation results related to this covariate in Section 6.
Two main aspects have to be addressed for a correct approach to the estimation ofequation (2): the panel structure of the data and the presence of spatial correlationof antibiotic consumption across regions. As suggested by Baltagi (2005), the use ofpanel data has some advantages compared to the use of pure time series or cross-sectional data. Panel data allow to control for individual heterogeneity and are more
8 Adding a positive scalar (k) to the values of a covariate (X) ensures that the function log(X +k)
is always de…ned. See Gujarati (1995) for details.
Density of physicians per 1,000 pop.
Table 2: Descriptive statistics of variables (years 2000-2008).
informative. Furthermore, they present more variability and less collinearity, andprovide more degrees of freedom and more e¢ cient estimates. Finally, they o¤er theadvantage that units are observed through time and this allows for a simpli…cationof economic aspects that, otherwise, would be more di¢ cult to study. The temporaldimension pertains to periodic observations of variables characterising cross-sectionalunits over time. Consequently, two sources of variation are identi…ed: the variationwithin units over time (within variation) and the variation across units (between vari-ation). These two sources of variation are di¤erently considered by the most widelyused econometric approaches to panel data: the pooled ordinary least-squares (OLS)model, the …xed-e¤ects (FE) model and the random-e¤ects (RE) model. Conversely,the spatial aspect pertains to the analysis of the e¤ects of the dependent variablebetween units (regions). This aspect will be taken into account by means of adequatespatial econometrics estimators included in the above models.
We estimate a FE model, which is a linear regression approach where the intercept
term varies cross-sectionally (over the individual units and/or over time). Thus,the common formulation of the model assumes that di¤erences across units can becaptured by di¤erences in intercept terms. To test the hypothesis of homogeneityin the constant terms across regions and time periods, we previously run an F -test.
The large F -test statistics suggests that a panel data approach via the FE estimatorwould give more e¢ cient estimates compared to the pooled OLS approach.
An alternative to the FE approach is the RE model, where the individual term
is a stochastic factor, independently and identically distributed across units. TheLagrange multiplier test (Breusch and Pagan, 1979) also indicates that the OLSmodel can be rejected in favour of the RE model. Moreover, using the Hausman testwe verify the hypothesis that the individual-speci…c error terms are uncorrelatedwith the explanatory variables, i.e. the RE estimator may be inconsistent. Since theHausman-test statistics is signi…cant at less than 5% level, we decide to focus on theconsistent …xed-e¤ects estimator (see Cameron and Trivedi, 2005, for more details).
Note also the relevance of the within variation in most of our covariates.
Concerning the spatial aspect, it is worth noting that regional antibiotic consump-
tion can be a¤ected by individuals’and physicians’attitudes towards antibiotics inadjacent regions. This externality problem can be taken into account by means of
adequate spatial econometrics estimators. There are two notable ways to introducespatial autocorrelation in regression models. These are the spatial-lag model and thespatial-error model. The former refers to a situation where antibiotic consumptionin one region is a¤ected by antibiotic consumption in nearby regions. The spatial-lagmodel is appropriate when there are spillover e¤ects from neighbouring regions. Thelatter model of spatial dependence focuses on the error term and assumes that errorterms in di¤erent regions are correlated. This kind of spatial dependence occurs ifthere are variables that are omitted from the regression model but do have an e¤ecton the dependent variable and are spatially correlated. It is the case, for instance,of random shocks spreading to neighbouring regions.
Both approaches requires the preliminary speci…cation of a matrix of spatial
weights (W ). This matrix contains information on the spatial association betweenobservational units. We construct a contiguity matrix indicating which regions sharea border.9 According to this proximity criterion, the elements of the spatial weightmatrix are 1 if location i is adjacent to location j, and zero otherwise. The stan-dardized matrix of spatial weights can then be used to test the presence of spatialautocorrelation.
We run two tests of spatial autocorrelation: the Moran’s I (Moran, 1948; Cli¤
and Ord, 1973; 1981) and Geary’s C statistics (Geary, 1954). Moran’s I statistic isa weighted correlation coe¢ cient formulated as a normalized quadratic form of thevariables tested for spatial correlation. Variables are standardised by subtractingthe sample mean and then de‡ated by the variance of the data (Anselin and Bera,1998). Values range from -1 to +1, where +1 indicates perfect positive correlation,0 implies no spatial correlation (provided the number of observations is large), and-1 indicates perfect negative correlation. Moran’s I values can then be transformedto Z-scores for statistical hypothesis testing. The Geary’s C statistics gives a valuebetween 0 and 2. The lower value indicates a strong positive spatial autocorrelation.
A value of 1 suggests that no spatial autocorrelation is present, whereas negativespatial association is suggested by a value greater than 1 (Goodchild, 1987). Thistest is inversely related to Moran’s I but is more sensitive to local rather than globalspatial autocorrelation.
The spatial-lag model can be de…ned as:
where DID is an N x1 vector of observations on antibiotic consumption per capita,with N = 160; W DID is the spatial lag of antibiotic consumption and
autoregressive parameter; X is the N xk matrix of explanatory variables, with k = 6;
is the vector of regression parameters and " is a vector of errors.
It is worth noting that spatial dependency is similar to having a lagged-dependent
variable as an explanatory variable. The spatial-lag model represents a suitable ap-proach to the study of spatial autocorrelation in antimicrobial consumption since it
9 In the case of Sardinia and Sicily the weights were assigned on the basis of the network of
maritime routes which links the two islands to the peninsula.
assumes that antibiotic use is characterised by consumption externalities as suggestedby the literature (Fingleton, 2003; Cabrer-Borrás and Serrano-Domingo, 2007). In-deed, antibiotics have a preventive e¤ect since their use may provide external bene…tsto other individuals, and consequently, reduce the need for consumption in neigh-bouring areas. However, antibiotics may also produce negative externalities sincetheir utilization may reduce antibiotic e¤ectiveness (increasing bacterial resistance)which may spread to other areas.
As an alternative to the spatial-lag model, we apply the spatial-error model. This
is more relevant than the spatial-lag approach when the distribution of residuals indi¤erent regions displays spatial correlation. Residuals may be spatially correlatedif aggregated shocks hit regional health authorities or there are unobservable riskfactors concentrated across the areas (Moscone and Knapp, 2005). This e¤ect maybe due, for instance, to exogenous bacterial resistance breakdown spreading acrossthe country.
The spatial-error model can be de…ned as:
is the spatial-autoregressive coe¢ cient and
assumed to be independently and identically distributed. Note from equation (5)that errors depend on the weighted average of errors in neighbouring regions.
Both spatial approaches have to deal with estimation bias. The multidirectional
nature of spatial dependence in the spatial-error model implies that generalized least-squares estimators are inconsistent. The spatial-lag model exhibits endogeneity thatcan be taken into account by instrumental variables or the general methods of mo-ments techniques, but should preferably be solved using an appropriate maximumlikelihood estimator (see Anselin 1988, for details).
In the context of panel data, …xed e¤ects can be included in the estimation of
equation (3), which leads to a …xed-e¤ects spatial-lag model (SLFE). A maximumlikelihood (ML) procedure can be used to estimate the model.10 Similarly, one canestimate a …xed-e¤ects spatial error-model (SEFE) (see e.g. Elhorst, 2003) usingequation (4) and (5). In both models, we use the lagged mortality rate instead ofthe mortality rate to tackle possible endogeneity related to the health status of thepopulation. Our estimations are carried out using the statistical software STATA(version 11).
Preliminary OLS regressions show an R2 adjusted of 0:77 (0:80 with a time trendor temporal dummy variables). The F test is 91:91 (92:53 and 49:28 with a time
1 0 The procedure developed by Pisati (2001) to investigate spatially correlated cross-sectional data
using maximum likelihood can be easily adapted to estimate a …xed-e¤ects spatial-lag model.
Table 3: Results of tests for spatial dependency.
trend and time dummies, respecitvely). This suggests that overall regressors have asigni…cant impact on the dependent variable. Moreover, the mean Variance In‡ationFactor is lower than 5. Finally, the Shapiro-Wilk test as well as the Jarque-Beratest for normality of errors cannot be rejected using the conventional 95% level ofsigni…cance.
Our initial tests also indicate that the OLS model can be rejected in favour of the
FE and the RE models. The Breusch and Pagan lagrangian multiplier test indicatesthat there are e¤ects other than those captured by the exogenous variables in OLSregressions. The F test that constant terms are homogeneous across regions and timeperiods is also rejected. Moreover, the Hausman test suggests that the FE approachshould be preferred to the RE approach. The Hausman test has a value of 74:44 anda p-value of 0:00. Consequently, the estimation results reported in Table 4 focus onthe FE model.
To account for possible cross-sectional dependence, we also estimated a FE model
using the Driscoll-Kraay correction to standard errors (Driscoll and Kraay, 1998).
Ignoring cross-sectional correlation in the estimation of panel models can lead toseverely biased statistical results. The statistical signi…cance of the coe¢ cients in ourregression slighly improves when the Driscoll-Kraay correction is applied. Finally, tocheck the robustness of our results and address possible endogeneity of populationhealth status we run separate regressions using the lag of mortality for infectionsand an instrumental variable approach using population density as instrument forinfections. The main results are unchanged.
Generally, only a few coe¢ cients are signi…cant. Nevertheless, the goodness of …t
of our preliminary estimation with OLS is not far from the results of previous analysisof antimicrobial use at regional level. Because we use a log-log functional form, we caninterpret coe¢ cients for income and price as elasticities. In the FE estimation, incomeelasticity is positive and highly signi…cant. The result suggests that regions withhigher levels of income, i.e. northern Italian regions, use more antibiotics comparedto lower income regions, ceteris paribus. Generally, a 1% increase in income increasesantibiotic consumption by 0:64%. Positive income e¤ects for antimicrobials are alsoobserved by Baye et al. (1997) using US data, and by Filippini et al. (2009a) usingSwiss data. According to the authors, one possible explanation for this low elasticityvalue is that the increasing concern over the e¤ects of bacterial resistance from the
1990s may have reduced income elasticity of outpatient antibiotic expenditure overtime. Another explanation is that individuals with higher income are more likely tosubstitute away antibiotics for alternative treatments when income increases.
As expected, copayment has a negative and signi…cant impact on consumption
( 0:015). Many studies suggest that copayments are e¤ective in reducing drug con-sumption at individual level (Freemantle and Bloor, 1996). Using data from Italy,Fiorio and Siciliani (2009) investigate the e¤ect of copayments on the demand forpharmaceuticals. They …nd that an increase in copayments by e1 reduces the percapita number of prescriptions by 4% and the per capita public pharmaceutical ex-penditures by 3:4%. Therefore, the e¤ect of a variation in the level of copaymentis not negligible. Our estimates are lower than those found by Contoyannis et al.
(2005), who investigate exogenous changes in the cost-sharing of prescription drugsin Canada (between
0:16). Own-price elasticities calculated by Rudholm
(2003) for three Swedish pharmaceutical submarkets between 1989 and 1996 are alsolower (between
Finally, we observe that the coe¢ cient of mortality rate for infectious diseases
( 0:01) is not signi…cant. This suggests that improvements in population healthstatus are not signi…cantly associated to increasing rates of antibiotic use.
The results of the two spatial models with …xed e¤ects (SLFE and SEFE) are
also reported in Table 4. For comparison purposes we run spatial random-e¤ectsestimations with maximum likelihood and observed that the sign and the signi…canceof all the coe¢ cients do not di¤er substantially from estimations with …xed e¤ects.
Table 3 summarises the results of two tests for spatial dependence: the Moran’s Itest and the Geary’s C test. In both cases, the null hypothesis is rejected, whichsuggests evidence of spatial autocorrelation in antibiotic use among Italian regions.
It is then advisable to extend our FE model to include interdependence of antibioticconsumption across regions by means of spatial models. In order to identify theappropriate form of spatial autocorrelation, we use two Lagrange multiplier (LM)tests and their robust versions. The LM test for a spatial lag and the LM test forspatially autoregressive errors both suggest the presence of spatial dependency (Table3). The LM test statistics for the spatially lagged dependent variable are alwayssigni…cant at the 0:1% level. Regarding the model with spatially autoregressiveerrors, only the robust LM test statistic is signi…cant at the 0:1% level. This leadsus to conclude that there is no clear evidence in favour of one speci…cation approachover the other. Estimation results for both spatial speci…cation approaches are thendiscussed and reported in Table 4.
We estimate the spatial-lag model and the spatial-error model taking unobserv-
able e¤ects into account by means of regional dummies. The maximul likelihoodprocedure for spatial analysis of cross-sectional data developed by Pisati (2001) hasbeen adapted to take …xed e¤ects into account. Estimations with a time trend andtime dummies have also been considered. Our results are robust to the inclusionof a time trend. The inclusion of temporal dummies reduces the signi…cance of thespatial e¤ects but the result is undermined by the large number of dummy variables(regional e¤ects and temporal e¤ects) and the small number of observations.
The estimates of the two spatial models are quite similar. Region dummies are
all highly signi…cant with a couple of exceptions. Income is highly signi…cant in bothmodels. Antibiotic copayment is signi…cant at less than 5%. The proportion of peopleaged above 64 is also signi…cant at less than 1% in the spatial-lag model and less than5% in the spatial-error model. This suggests that elderly individuals are less likelyto use outpatients antibiotics compared to younger individuals. According to theliterature there is a U -shaped relationship between health care spending and age (DiMatteo, 2005). Young and elderly individuals generally use more health services thanthe mid-age population. In the case of outpatient antibiotics the relationship seemsto be reinversed maybe because individuals in the labor force have higher opportunitycost of time and tend to shorten time to recover by means of drug therapy.
As for the spatial coe¢ cients, these are signi…cant in the spatial-lag and the
spatial-error models at 5% and 1% level respectively. One could argue that oneof these two models captures the dynamics of the externalities involved (bacterialresistance and prevention from infections) better than the other. This would implythat the spillover process is either deterministic or similar to a random shock. Oneresult of our analysis is that the spatial-lag parameter in the spatial-lag model ( ) isless signi…cant than the spatial autocorrelation coe¢ cient in the spatial-error model( ). This may suggest that random shocks related to the dynamics of infections maybetter explain spatial interactions between neighbouring areas in the consumption ofantibiotics.
Most of the empirical evidence on socioeconomic determinants of antibiotic consump-tion is based on cross-sectional data and limited to few countries (e.g. Switzerland,Germany, Israel and Hungary). In particular, evidence lacks from countries with aNational Health Service and substantial decentralization of health care provision toregional health authorities. The cross-sectional approach has some drawbacks suchas the inability to solve the problem of omitted variables. This is even more relevantwhen only limited data are available for important factors, such as bacterial resis-tance to antimicrobials. A further limitation is represented by the need to impose fullregional homogeneity in the parameters of the random process that describes the useof antibiotics. To overcome these problems, we analysed socioeconomic determinantsof antibiotic consumption by means of panel data from a new country (Italy).
The consumption of antibiotics cannot be regarded as independently generated
within regions because of possible spillover e¤ects. Antibiotics may reduce the risk ofinfections in neighbouring areas (positive externality) and may reduce the e¤ective-ness of treatment because of bacterial resistance spreading (negative externality). Asa consequence, standard estimation procedures employed in many empirical studiescan lead to bias and ine¢ ciency in the estimates. Our approach allowed to considerspatial e¤ects across regions. We captured these e¤ects by means of a spatial-lagmodel and a spatial-error model.
We found some evidence of spatial autocorrelation in the use of antibiotics across
Italian regions. This suggests that regional policies (e.g. public campaigns) aimedat increasing e¢ ciency in antibiotic consumption and controlling bacterial resistancemay not be independent and could be in‡uenced by policy makers in neighbouringregions. There will be scope for a strategic and coordinated view of regional policiestowards the use of antibiotics.
Albricht, W., Monnet, D.L. and Harbarth S. (2004) Antibiotic selection pressureand resistance in Streptococcus pneumoniae and Streptococcus pyogenes, EmergingInfectious Diseases, 10, 514-17.
Anselin, L. (1988) Spatial econometrics: Methods and models, Kluwer, Dordrecht,the Netherlands.
Anselin, L. and Bera, A. (1998) Spatial dependence in linear regression models withan introduction to spatial econometrics, in Handbook of Applied Economic Statistics,(Eds) A. Ullah and D.E.A. Giles, Marcel Dekker, New York, pp. 237-89.
Atella, V., Hassell, K, Schafheutle, E., Weiss M.C. and Noyce P.R. (2003) Cost tothe patient or cost to the healthcare system? Which one matters the most for GPprescribing decisions? A UK-Italy comparison, CEIS Tor Vergata Research PapersNo. 1.
Baltagi, B. (2005) Econometric analysis of panel data, 3rd edn, Wiley, Chichester.
Baye, M.R., Maness, R. and Wiggins, S.N. (1997) Demand systems and the truesubindex of the cost of living for pharmaceuticals, Applied Economics, 29, 1179-90.
Bretteville-Jensen, A.L. (2006) Drug demand –initiation, continuation and quitting–, De Economist, 154, 491-516.
Breusch, T. and Pagan, A. (1979) A simple test of heteroskedasticity and randomcoe¢ cient variation, Econometrica, 47, 1287-94.
Cabrer-Borrás, B. and Serrano-Domingo, G. (2007) Innovation and R&D spillovere¤ects in Spanish regions: A spatial approach, Research Policy, 36, 1357-71.
Cameron, A. and Trivedi, P. (2005) Microeconometrics. Methods and Applications,Cambridge, Cambridge University Press.
Cli¤, A. and Ord, J. 1973 Spatial Autocorrelation, London, Pion.
Cli¤, A.D. and Ord, J.K. (1981) Spatial Processes: Models and Applications, London,Pion.
Contoyannis, P., Hurley, J., Grootendorst, P., Jeon, S.H. and Tamblyn, R. (2005)Estimating the price elasticity of expenditure for prescription drugs in the presence ofnon-linear price schedules: an illustration from Quebec, Canada. Health Economics,14, 909-23.
Costa-Font, J., Kanavos, P. and Rovira, J. (2007) Determinants of out-of-pocketpharmaceutical expenditure and access to drugs in Catalonia, Applied Economics,39, 541-551.
Costa-Font, J., Moscone, F. (2008) The impact of decentralization and inter-territorialinteractions on Spanish health expenditure, Empirical Economics, 34, 167-184.
Di Matteo, L. (2005) The macro determinants of health expenditure in the UnitedStates and Canada: assessing the impact of income, age distribution and time, HealthPolicy , 71, 23-42.
Driscoll, J., Kraay, A.C. (1998) Consistent covariance matrix estimation with spa-tially dependent data, Review of Economics and Statistics, 80, 549-560.
Elhorst, J. (2003) Speci…cation and estimation of spatial panel data Models, Inter-national Regional Science Review, 26, 244-68.
Elseviers, M., Ferech, M., Vander Stichele, R.H. and Goossens, H. (2007) Antibioticuse in ambulatory care in Europe (ESAC data 1997-2002): trends, regional di¤erencesand seasonal ‡uctuations, Pharmacoepidemiology and Drug Safety, 16, 115-123.
European Commission. (2010) Antimicrobial Resistance, Eurobarometer 338/Wave72.5 –TNS Opinion & Social, Luxembourg.
Ferech, M., Coenen, S., Dvorakova, K., Hendrickx, E., Suetens, C. and Goossens,H. (2006) European Surveillance of Antimicrobial Consumption (ESAC): outpatientpenicillin use in Europe, Journal of Antimicrobial Chemotherapy, 58, 408-12.
Filippini, M., Masiero, G. and Moschetti, K. (2006) Socioeconomic determinants ofregional di¤erences in outpatient consumption: Evidence from Switzerland, HealthPolicy, 78, 77-92.
Filippini, M., Masiero, G. and Moschetti, K. (2009a) Small area variations and wel-fare loss in the use of outpatient antibiotics, Health Economics, Policy and Law, 4,55-77.
Filippini, M., Masiero, G. and Moschetti, K. (2009b) Regional consumption of an-tibiotics: a demand system approach, Economic Modelling, 26, 1389-97.
Fingleton, B. (2003) Externalities, economic geography, and spatial econometrics:conceptual and modeling developments, International Regional Science Review, 26,197-207.
Fiorio, C. and Siciliani, L. (2010) Co-payments and the demand for pharmaceuticals:Evidence from Italy, Economic Modelling, 27, 835-41.
Freemantle, N. and Bloor, K. (1996) Lessons from international experience in con-trolling pharmaceutical expenditure I: in‡uencing patients, British Medical Journal,312, 1469-71.
Geary, R. (1954) The contiguity ratio and statistical mapping, The IncorporatedStatistician, 5, 115-45.
Goodchild, M.F. (1987) A spatial analytical perspective on geographical informationsystems, International Journal of Geographical Information Systems, 1, 327-34.
Gujarati, D.N. (1995) Basic Econometrics, 3rd edn, McGraw-Hill, New York.
Habitch, J., Kiivet, R.A., Habicht, T. and Kunst, A.E. (2009) Social inequalities inthe use of health care services after 8 years of health care reforms – a comparativestudy of the Baltic countries, International Journal of Public Health, 54, 1-10.
Huttner, B., Goossens, H., Verheij, T. and Harbarth, S. on behalf of the CHAMPconsortium. (2010) Characteristics and outcomes of public campaigns aimed at im-proving the use of antibiotics in outpatients in high-income countries, Lancet Infec-tious Diseases, 10, 17-31.
Kern, W.V., de With, K., Nink, K., Steib-Bauert, M. and Schröder, H. (2006) Re-gional variation in outpatient antibiotic prescribing in Germany, Infection, 34, 269-73.
Masiero, G., Filippini, M., Ferech, M. and Goossens, H. (2010) Socioeconomic de-terminants of outpatient antibiotic use in Europe, International Journal of PublicHealth, 55, 469-78.
Matuz, M., Benko, R., Doro, P., Haidu, E., Nagy, G., Nagy, E., Monnet, D.L.
and Soos, G. (2005) Regional variations in community consumption of antibiotics in
Hungary, 1996-2003, British Journal of Clinical Pharmacology, 61, 96-100.
Mera, R.M., Miller, L.A. and White A. (2006) Antibacterial use and Streptococcuspneumoniae penicillin resistance: a temporal relationship model, Microbial DrugResistance, 12, 158-63.
Monnet, D.L., Mölstad, S. and Cars, O. (2004) De…ned daily doses of antimicro-bials re‡ect antimicrobial prescriptions in ambulatory care, Journal of AntimicrobialChemotherapy, 53,1109-11.
Monroe, S. and Polk, R. (2000) Antimicrobial use and bacterial resistance, CurrentOpinion in Microbiology, 3, 496-501.
Moran, P. (1948) The interpretation of statistical maps, Journal of the Royal Statis-tical Society Series B, 10, 243-51.
Moscone, F. and Knapp, M. (2005) Exploring the spatial pattern of mental healthExpenditure, Journal of Mental Health Policy and Economics, 8, 205-17.
Nitzan, O., Low, M., Lavi, I., Hammerman, A., Klang, S. and Raz, R. (2010) Vari-ability in outpatient antimicrobial consumption in Israel, Infection, 38, 12-18.
Pisati, M. (2001) Tools for spatial data analysis, Stata Technical Bulletin stb-60.
Revelli, F. (2001) Spatial patterns in local taxation: tax mimicking or error mimick-ing?, Applied Economics, 33, 1101-07.
Rudholm, N. (2003) Competition and substitutability in the Swedish pharmaceuticalsmarket, Applied Economics, 35, 1609-17.
Neutral Citation Number: 2013 EWHC 2125 (Ch) IN THE HIGH COURT OF JUSTICE CHANCERY DIVISION PROBATE In the estate of LOUISA ANN ASHKETTLE deceased Christopher Pymont QC - - - - - - - - - - - - - - - - - - - - - Between : (1) ROBERT MICHAEL ASHKETTLE Claimants (2) DENNIS ROBERT ASHKETTLE ROSALIND PATRICIA ANN GWINNETT Defendant - - - - - - - - - - - - - - - - -
MOST COMMONLY PRESCRIBED DRUGS (Preferred Drug List) EFFECTIVE jaNUaRY 1, 2007 The Blue Cross and Blue Shield of Texas most commonly If you are currently taking a drug that is not shown on this list, prescribed preferred drugs are listed below. This list does not cal Customer Service at the number located on the back of your include all of the preferred drugs that are included in yo