Is there room for containing healthcare costs? An analysis of regional spending differentials in Italy

Abstract

This work aims at identifying the determinants of health spending differentials among Italian regions and at highlighting potential margins for savings. The analysis exploits a data set for the 21 Italian regions and autonomous provinces starting in the early 1990s and ending in 2006. After controlling for standard healthcare demand indicators, remaining spending differentials are found to be significant, and they appear to be associated with differences in the degree of appropriateness of treatments, health sector supply structure and social capital indicators. In general, higher regional expenditure does not appear to be associated with better reported or perceived quality in health services. In the regions that display poorer performances, inefficiencies appear not to be uniformly distributed among expenditure items. Overall, results suggest that savings could be achieved without reducing the amount of services provided to citizens. This seems particularly important given the expected rise in spending associated with the forecasted demographic developments.

Appendices

Appendix 1: Robustness checks

To check the robustness of our estimates, we run a series of exercises. We estimate model 1 using alternative estimation techniques. We start by considering a stochastic frontier model⁴⁵ of the type:

$$ c_{it} = f(x_{it} )\xi_{it} $$

(9)

where $\xi_{it} = 1$ for the best unit in the sample and $\xi_{it} > 1$ for the others.

We specify the efficient expenditure function as

$$ f(x_{it} ) = e^{{{\text{cons}} + {\mathbf{x^{\prime}}}_{it} \beta }} $$

(10)

and hence we can write the estimating equation as

$$ \ln c_{it} = {\text{cons}} + {\mathbf{x^{\prime}}}_{it} \beta + u_{it} + \nu_{it} $$

(11)

where $\nu_{it} {\text{ iid }}N(0,\sigma_{\nu }^{2} )$ is the error term and $u_{it} {\text{ iid }}N^{ + } (\mu ,\sigma_{u}^{2} )$ captures each region’s distance with respect to the efficient frontier⁴⁶:

$$ \xi_{it} = e^{{u_{it} }} $$

(12)

Initially, we assume that the stochastic frontier is invariant over time.⁴⁷ We then consider a time varying version à la Battese and Coelli [1]:

$$ u_{it} = e^{ - \eta (t - T)} u_{i}\quad{\text{with}}\quad u_{it} {\text{iid}}N^{ + } (\mu ,\sigma_{u}^{2} ) $$

(13)

where the distance from minimum cost can be decomposed into two factors, a given level of inefficiency u _i characterising each region and a time varying component, which is a function of time and a decay factor $\eta$.

We also estimate a fixed effect model with time varying regional effects. In particular, we derive an estimating equation equivalent to the time varying stochastic frontier described above (13). In this case the expenditure equation becomes⁴⁸:$\ln c_{it} = {\mathbf{x^{\prime}}}_{it} \beta + e^{ - \eta t} \phi_{i} + \varepsilon_{it}$ with

$$ \phi_{i} = [e^{\eta T} \alpha_{i} ] $$

(14)

Using the approximation

$$ e^{ - \eta t} \cong 1 - \eta t + \eta^{2} \frac{{t^{2} }}{2} $$

(15)

we derive the following estimating equation⁴⁹:

$$ \ln c_{it} = {\mathbf{x^{\prime}}}_{it} \beta + \phi_{i} - \eta t\phi_{i} + \eta^{2} \frac{{t^{2} \phi_{i} }}{2} + \varepsilon_{it} $$

(16)

From the estimates of (16) we can recover the parameters we are interested in: the regional (time invariant) effects α _i and the decay factor η. In both time-varying models the estimated decay factor is negative and significant. This might reflect a common increasing pattern in health expenditure that could be due to the impact of technological developments in the production of healthcare, in the organisation at national level of the health system or in the structure of preferences.

Appendix 2: Indicator for the distribution of inefficiency among expenditure items

The indicator $\vartheta_{i}^{h}$ can be rewritten as follows: $\theta_{i}^{h} = \frac{{s_{i}^{h} \tilde{\xi }_{i}^{h} - s_{i}^{h} }}{{\xi_{i} - \xi_{{i{\text{ ref}}}} }} - s_{i}^{h}$ or

$$ \theta_{i}^{h} = s_{i}^{h} \frac{{\tilde{\xi }_{i}^{h} - \xi_{i} }}{{\xi_{i} - \xi_{{i{\text{ ref}}}} }} $$

(17)

when $\tilde{\xi }_{i}^{h} = \xi_{i} \, \forall h$ then $\theta_{i}^{h} = 0 \, \forall h$. All the items have the same level of inefficiency. The distortion might be very big but it is uniformly distributed among the expenditure items. On the other hand, consider a case in which all the first h–1 items are characterised by ‘full efficiency’ so that $\tilde{\xi }_{i}^{h} = 1 \, \forall h = 1, \ldots ,H - 1$. In this case with $\xi_{{i{\text{ ref}}}} = \sum\limits_{h = 1}^{H} {s_{i}^{h} = 1}$, we will have $\theta_{i}^{h} = - s_{i}^{h} \, \forall h = 1, \ldots ,H - 1$ while the overall inefficiency will reflect the distortion in the last expenditure item: $\theta_{i}^{H}$ will be $> 0 > \theta_{i}^{h} \, \forall h = 1, \ldots ,H - 1$. Given (7), we will in fact have that: $\xi_{i} = \left( {\sum\limits_{h = 1}^{H - 1} {s_{i}^{h} } } \right) + s_{i}^{H} \tilde{\xi }_{i}^{H}$. Finally it should be noted that for items characterised by $\tilde{\xi }_{i}^{h} > \xi_{i} , { }\theta_{i}^{h} \,$will be positive, while for $\tilde{\xi }_{i}^{h} < \xi_{i} , { }\theta_{i}^{h} \,$will be negative. This means that a large distortion in a relatively small spending item might be compensated by a relatively smaller distortion in a spending item that accounts for a large fraction of total outlays.