1. JECH Online First, published on September 29, 2012 as 10.1136/jech-2011-200640Research reportThe effect of healthcare delivery privatisation onavoidable mortality: longitudinal cross-regional resultsfrom Italy, 1993e2003Cecilia Quercioli,1 Gabriele Messina,1 Sanjay Basu,2,3,4 Martin McKee,4 Nicola Nante,1David Stuckler4,5ABSTRACTBackground During the 1990s, Italy privatiseda significant portion of its healthcare delivery. Theauthors compared the effectiveness of private and publicsector healthcare delivery in reducing avoidable mortality(deaths that should not occur in the presence of effectivemedical care).Methods The authors calculated the average rate ofchange in age-standardised avoidable mortality rates in19 of Italys regions from 1993 to 2003. Multivariateregression models were used to analyse the relationshipbetween rates of change in avoidable mortality andlevels of spending on public versus private healthcaredelivery, controlling for potential demographic andeconomic confounders.Results Greater spending on public delivery of healthservices corresponded to faster reductions in avoidablemortality rates. Each V100 additional public spending percapita on NHS delivery was independently associatedwith a 1.47% reduction in the rate of avoidable mortality(p0.003). In contrast, spending on private sectorservices had no statistically significant effect onavoidable mortality rates (p0.557). A higher percentageof spending on private sector delivery was associatedwith higher rates of avoidable mortality (p0.002). Theauthors found that neither public nor private sectordelivery spending was significantly associated withnon-avoidable mortality rates, plausibly because non-avoidablemortality is insensitive to healthcare services.Conclusion Public spending was significantly associatedwith reductions in avoidable mortality rates over time,while greater private sector spending was not at theregional level in Italy.INTRODUCTIONItalys healthcare system underwent significantchanges in the early 1990s. The Servizio SanitarioNazionale (National Health Service, abbreviatedSSN) was created in 1978 as a nationwide servicefunded predominantly by general taxation, basedlargely on state provision and free at the point ofuse.1 In the 1990s, continuous increases in health-carecosts and perceived inefficiency created pres-sureto introduce major reforms, including increasesin the role of the private sector in provision anddecentralisation of health policy responsibilities toregional administrations.2 3 To varying degrees,healthcare delivery was shifted from the publicsystem to the private sector.4 5 Financing for theservice continued to be public, coming mainly fromtax revenues, but over time, the public share oftotal finance decreased from 80.5% of the 1980 to72.6% of the 2000.6 The creation of internalmarkets and competition within regions were seenas a means to reduce waste and improve theresponsiveness of healthcare providers to patientsdemands.4 7How would these significant reforms be expectedto affect healthcare performance? The existingliterature on ownership of healthcare delivery iscomplex and suffers from many limitations.8e10 Itis dominated by data from American hospitals andmuch uses Medicare data, which cover only thoseaged 65 years or older. Many of the papers arereanalyses of the same data sets. A 2002 systematicreview concluded that for-profit hospitals producedsignificantly worse outcomes than not-for-profitones.11 However, the authors of a recent meta-regressionthat sought to address the limitations ofthe earlier work were more guarded and concludedsimply that studies representative of the USA asa whole tend to find lower quality among FPs (for-profits)than private non-profits.12 Limited researchexists from outside the USA. A study in one prov-inceof China found no difference in mortalityaccording to hospital ownership,13 while anAustralian study found better outcomes followingmyocardial infarction in private compared withpublic hospitals, which seemed to be associatedwith the considerably more intensive investigationand treatment in the latter.14 More recently, severalstudies have been undertaken in the UK looking atthe analogous issue of competition in provision ofcare within a largely publicly owned system andhealth outcomes (specifically mortality frommyocardial infarction). These studies purport toshow an association. Although cited extensively bypoliticians in support of radical reforms to theEnglish NHS, it remains highly controversial, withcritics noting considerable methodological prob-lemswith the data, the measure of competition andthe outcome used, the plausibility of the findings,and that the findings display the inverse of theauthors interpretationdthat outcomes worsenedin non-competitive areas15 and the authors havepointed out that Researchers have little influenceover how politicians use their academic work,16in particular noting that price competition isassociated with worse outcomes.17There are two main benefits to studyingthe Italian case of privatisation of healthcaredelivery. First, Italy essentially undertook a quasi-naturalexperiment with privatisation and decen-tralisation.18 Such measures are currently being1Dipartimento di Fisiopatologia,Medicina Sperimentale e Sanita`Pubblica, Universita` degli Studidi Siena, Siena, Italy2Department of Medicine,University of California SanFrancisco, San Francisco,California, USA3Division of General InternalMedicine, San FranciscoGeneral Hospital, San Francisco,California, USA4Department of Public Healthand Policy, London School ofHygiene & Tropical Medicine,London, UK5Department of Sociology,University of Cambridge,Cambridge, UKCorrespondence toDr Cecilia Quercioli, Departmentof Public Health, University ofSiena, Via Aldo Moro 2, 53100Siena, Italy; quercioli2@unisi.itAccepted 26 June 2012Quercioli C, Messina G, Basu S, et al. J Epidemiol Community Health (2012). doi:10.1136/jech-2011-200640 1 of 7Copyright Article author (or their employer) 2012. Produced by BMJ Publishing Group Ltd under licence.

2. Research reportconsidered by England and Greece, among others. Because of thevaried degree of privatisation among different regions during theperiod 1993e2003, with almost no increase in private sectordelivery in Puglia as a fraction of total delivery but a rise of 9percentage points in Lombardy and Veneto regions (figure 1), theoutcomes between regions can be directly compared aftercontrolling for region-specific variables. Second, the economiccontext of existing studies differs from that in Italy. Forexample, Medicare (the Australian health insurance system)pays only a contribution towards care in private hospitals withpatients paying the rest, either directly or via insurance. Hence,the funding for private care in Australia is rather more generousthat publicly provided care. It is plausible that, in Italy, givenfixed funding per case (using the Diagnosis Related Groups,a prospect payment system19), the need to extract profits andmoney to repay loans from private facilities on the one handcould be a incentive to improve efficiency but on the other handcould determine a reduction in the amount available for deliv-eringhigh-quality patient care. The Italian privatisation exper-imentpotentially has important lessons for other countriesFigure 1 Variations across Italysregions. Panel A: change in privatesector delivery, 2003 vs 1993. Panel B:average rate of reduction in avoidablemortality, 1993e2003.2 of 7 Quercioli C, Messina G, Basu S, et al. J Epidemiol Community Health (2012). doi:10.1136/jech-2011-200640 3. considering similar reforms as they seek to address budget defi-citsfollowing the 2008 financial crisis.In this study, we examine whether the share of privatedelivery of healthcare in Italy affected each regions progress inone indicator of health system performance: avoidable mortality.As defined by Rutstein et al,20 avoidable mortality comprisesdeaths that should not occur in the presence of effectivemedical care.METHODSPopulation numbers by age, gender and region and number ofdeaths by age, gender, region and cause of death in the period1993e2003 were extracted from computer files of the ItalianBureau of Statistics (ISTAT). Mortality data were classified usingthe 9th Revision of the International Classification of Diseases(ICD-9) for the years 1993e2002 and using the 10th RevisionICD-10 for the year 2003. We adjusted the number of deaths in2003 using the bridge coding system ICD-9/ICD-10 developedby the ISTAT.21 The period 1993e2003 was chosen because itcovers the years of the main SSN reforms and because at thetime of the analysis regional mortality data for 2004 and 2005were not yet available. Avoidable mortality was categorisedfollowing the methods of Rutstein et al,20 updated by Nolte andMcKee.22 Types of death included in the mortality statistic arelisted in box 1. To avoid potential confounding by age and toensure a fair comparison across avoidable and non-avoidablemortality clusters, both mortality rates were age standardised tothe European standard population.23Regional data on healthcare expenditure and on socioeco-nomic-lifestyle indicators (Gross Domestic Product per capita,percentage of people with university degree, and smoking rates)were obtained from the Health for AlldItaly database and fromdatabases/publications of ISTAT and the Italian Ministry forHealth.24 25Private delivery was calculated as the sum of the out-of-pockethealth expenditure per capita and the health expenditureper capita of the SSN for services delivered by private providers(including outpatient specialist care and hospital care). Publicdelivery was calculated as the sum of the public expenditure forservices provided directly by the SSN (per capita), the pharma-ceuticalexpenditure provided by SSN (per capita), the expen-ditureper capita for primary care (General Practitioners),Research reportexpenditure for public outpatient specialist care but notincluding administrative expenditure of SSN (per capita). We didnot include administrative expenditure because this is an over-headand does not directly affect the care received by patients;however, the results were found not to have been qualitativelychanged by this step. Some regions may have more or lessoverhead costs, depending on the efficiency of care organisation,which could consequently dilute the effects of public caredelivery, which we sought to evaluate.Given that the ability of a health system to reduce deathsfrom any cause depends on numerous complex regional, andoften historical, factors ranging from availability of healthcareproviders to social determinants of mortality, it is more relevantto look at rates of change in regional mortality rather thancompare absolute levels of mortality among regions. This isespecially important in Italy, given the geographical, economicand cultural diversity of the country, and because each regionbegan privatising healthcare delivery at different initial levels ofavoidable mortality. Consequently, we have evaluated the rate ofprogress in reducing avoidable mortality over time by holdingbetween-region variations constant, using a fixed effectsmodel that accounts for the proportion of differences in rates ofmortality change that are due to baseline differences betweenregions.26 The fixed effects framework offers similar advantagesto those with a difference-in-difference approach in so far as itevaluates the variations within a region attributable to thepolicy; the difference-in-difference approach is a special caseinvolving a binary treatment variable where a policy is turnedon at the time of implementation. Importantly, both difference-in-difference and fixed effects models hold constant between-groupheterogeneity, in the case of the former by differencingand in the latter by time demeaning the data.To check the specificity of our findings, we also tested theassociation of public and private delivery with non-avoidablemortality rates, which we hypothesised would not be associatedwith changes in the nature of healthcare delivery. Using non-avoidablemortality rates also has the important advantage ofdetecting potential confounding; if rates of public delivery acrossregions were confounded by an unobserved factor, which unre-latedto the health system, such as lower tobacco use or betterhealth status, they would manifest as a significant associationwith non-avoidable mortality rates.Thus, our initial statistical model was as follows:Avoidable Mortality Deliveryi;ta bPublic i;t bPrivate Deliveryi;t t mi 3i;t(1)Here, i is region and t is year. All the economic variables aremeasured in euro current prices. m is the region dummy variable(controlling for region-specific differences), t is a time trend and3 is the error term. Both public and private deliveries wereincluded in the model to assess their independent effects. Wheretwo covariates are highly correlated, such an modelling approachcan potentially give rise to unstable estimates (multi-collinearity).There is speculation that public and privatedelivery may be correlated either negatively (crowd out) orpositively (crowd in). However, using the most recent year ofdata, we found evidence for neither of these possibilities(r0.028, p0.91). Additionally, in the second step, we investi-gatedthe effect of greater private sector share of delivery,measured as private delivery as a fraction of total delivery. Ineach model, we included a time trend to adjust for averagereductions in mortality rates that applied evenly to ItalysBox 1 Categories of avoidable mortalityAges 0e14: intestinal infections, whooping cough, measles(1e14) and all respiratory diseases except pneumonia andinfluenza (1e14).Ages 0e44: malignant neoplasm of cervix uteri and body ofcervix, leukaemia.Ages 0e49: diabetes and asthma.Ages 0e74: tuberculosis; other infections (diphtheria, tetanus,septicaemia and poliomyelitis); syphilis; malignant neoplasm ofcolon and rectum, skin, breast, cervix uteri and testis; Hodgkinsdisease; diseases of the thyroid; epilepsy; deficiency anaemia;chronic rheumatic heart disease; hypertensive disease; ischaemicheart disease (half of deaths); cerebrovascular disease; influenza;pneumonia; peptic ulcer; appendicitis; abdominal hernia; choleli-thiasisand cholecystitis; nephritis and nephrosis; benign pros-tatichyperplasia; misadventures to patients; maternal death;congenital cardiovascular and digestive nomalies; and perinataldeaths, all causes, except stillbirths.Quercioli C, Messina G, Basu S, et al. J Epidemiol Community Health (2012). doi:10.1136/jech-2011-200640 3 of 7 4. Research reportregions but were unrelated to changes to variations in healthsystem delivery. In a series of further robustness checks, weincluded adjustments for additional time-varying factors,including gross domestic product per capita (a measure ofincome) measured in euro constant price, the percentage ofpopulation with tertiary education and the percentage of theadult population using tobacco.As a diagnostic test, we removed regions that had large year-to-year fluctuations in avoidable mortality of >20%, whichwould be implausible in the absence of external shocks. Thisonly occurred in Aosta Valley, which has a very small populationof about 120 000 inhabitants, and as consequence a very smallnumber of deaths resulting in statistically non-sensical vari-ability.Thus, we excluded it from the model because its largefluctuations in avoidable mortality would compose a substantialfraction of overall variation in the data set in percentage changeterms and could thus qualitatively change our findings.Data were analysed using STATA V.10.2 StataCorp LP, CollegeStation, Texas 77845 USA. Standard errors were clustered byeach region to reflect non-independence of sampling. Modelcoefficients were presented as semi-elasticities (ie, as percentagechanges) to facilitate interpretation.Figure 2 Public and private healthcaredelivery and progress in reducingavoidable mortality, 1993e2003.Panel A: relationship between averagelevels of spending on public healthcaredelivery and average rate of reduction inavoidable mortality. Panel B:relationship between average levels ofspending on private healthcare deliveryand average rate of reduction inavoidable mortality.4 of 7 Quercioli C, Messina G, Basu S, et al. J Epidemiol Community Health (2012). doi:10.1136/jech-2011-200640 5. RESULTSFirst, we evaluated the relationship between each regions overallprogress in reducing age-standardised avoidable mortality ratesand average levels of spending on public and private delivery percapita (figure 2). Increased public spending on SSN delivery wascorrelated with an increase in the rate of decline in avoidablemortality between 1993 and 2003 (r0.33, figure 2, panel A).This association was weaker for spending in the private sector(r0.01, figure 2, panel B). Neither public nor private sources ofdelivery were associated with reductions in non-amenablemortality (rSSN0.13, rprivate0.17, respectively).iHowever, these correlations only allow for a comparison of 19data points since the progress over time (the slope) is based onan estimate derived from the longitudinal data, and both publicand private spending are correlated with each other and withGross Domestic Product (GDP). Our statistical model made itpossible to assess the independent associations of public andprivate spending with avoidable and non-avoidable mortalityrates over time, while correcting for potential bias from the non-randomimplementation of privatisation across Italys regions.Table 1 shows the results of our longitudinal fixed-effectstatistical models. We found that each V100 additional spendingon public sector services was associated with 1.47% lower ratesof avoidable mortality over the 1993e2003 period (p0.003). Incontrast, private sector spending had no statistically significanteffect on avoidable mortality during the period (p0.557). Takentogether, these results suggest that greater reliance on privatespending is not associated with an increase in mortality buta slower rate of progress in reducing avoidable morality thanpublic sector spending, an observation consistent with theunadjusted scatter plots in figure 2. We found that a higherpercentage of spending on private sector delivery was associatedwith higher rates of avoidable mortality (0.34%, p