Catastrophic health care payments. Measuring incidence and intensity of catastrophic impact

Transcription

1 Catastrophic health care payments Introduction One conception of fairness in payments for health care is that households ought not be required to spend more than a given fraction (say z) of their income on health care in any given period, and that spending in excess of this threshold can be labelled catastrophic [1]. The catastrophic label mainly refers to the fact that falling ill can induce often sizeable and unpredictable shocks a household s living standards. Clearly, the extent to which illness shocks really result in catastrophic economic consequences for households depends not only on medical care costs, but obviously also on any effects from reduced labor supply and productivity, and on the extent to which households are able to smooth their consumption over several periods by borrowing and lending mechanisms [2]. Here we focus only on the medical care expenses incurred and on the one period consequences (typically one year). Catastrophic payments defined The two key variables underlying the approach are: the health care payment variable whose catastrophic impact one wishes to assess; and a variable capturing the living standards of the persons to whom the payment applies. Other Technical Notes discuss the measurement of key health sector variables and the measurement of household living standards. The data have to be at the household level (e.g. household survey data), in which case one has, for each household, values of both the payments variable T and the living standards variable. Living standards may be measured by total per capita (or equivalent) expenditure, x, or alternatively in terms of a measure which more closely reflects ability to pay or discretionary expenditure, y=x-d(x), where D(x) indicates necessary or non-discretionary expenditure on items such as food. Catastrophic payments are then defined as those expenses for which the fraction T/x or T/y exceeds a pre-specified threshold z. The appropriate level of z is, to some extent, arbitrary and would clearly depend on whether living standards were measured by x or y. A health spending fraction of 20% of all expenditure may be considered catastrophic if the denominator refers to total expenditure but it may not be regarded as catastrophic if the denominator only includes the discretionary (e.g. non-food) component of all expenditure. Measuring incidence and intensity of catastrophic impact One straightforward way to summarize the extent to which a given sample of individuals has been exposed to catastrophic expenses is the number (or fraction) of individuals whose health care costs as a proportion of income exceed the threshold. The horizontal axis in Figure 1 shows the cumulative share of the sample, ordered by the ratio T/x, beginning with individuals with the largest ratio. Reading off this parade at the threshold z, one obtains the fraction H of the sample whose expenditures as a proportion of their income exceed the threshold z. Thisisthecatastrophic payment headcount. Define an indicator, which equals 1ifT i /x i >z and zero otherwise. Then the headcount is given by: (1) 1 N N i= 1 H =, i where N isthesamplesize. This measure does not reflect the amount by which households exceed the threshold. Another measure, the catastrophic payment gap (or excess), captures the average degree by which payments (as a proportion of income) exceed the threshold z. Define the excess or overshoot as Oi = i(( Ti / xi) z),i.e. the amount by which the payment fraction T i /x i exceeds the catastrophic threshold z. Then the gap is simply the average overshoot given by Catastrophic payments Page 1

2 (2) G 1 N N i= 1 = O, i In Figure 1, G indicates the area under the payment share curve but above the threshold level. It is clear that while H only captures the incidence of any catastrophes occurring, G also captures the intensity of the occurrence. They are related through the mean positive gap which is defined as G (3) MPG =. H Because this implies G = H. MPG, it means that the overall mean catastrophic gap equals the fraction with catastrophic payments times the mean positive gap. Obviously, all of the above measures can be computed using either x or y as denominator. Figure 1: Catastrophic payments as share of pre-payment income, by cumulative % of population, ranked by decreasing payment fraction Payments as share of income z Total catastrophic excess G Proportion H exceeding threshold 0% 100% Cum % of pop, ranked by decreasing payment fraction The various measures are illustrated using data taken from the 1993 Vietnamese Level of Living Survey in Box 1. Catastrophic payments Page 2

3 Box 1: Out-of-pocket payment shares for Vietnam, 1993 Table 1 presents some results on the catastrophic impact of out-of-pocket payments for health care in Vietnam in 1993 and illustrates its sensitivity to the choice of expenditure variable in the denominator and to different threshold level. Raising the level from 5 to 15% makes the proportion of the population with catastrophic payments drop from over 38% to below 10%, while the mean gap drops from just below 3% of expenditure to under 1%. Obviously, both the incidence and intensity is larger when catastrophe is defined as a share of discretionary, non-food expenditure per capita. This is also illustrated graphically in Figure 2, which shows the oop share curves for both definitions. Obviously, catastrophic shares are more prevalent when the narrower definition of (discretionary) expenditure is used. For instance, 20% of all Vietnamese households spent almost a quarter of their non-food spending on health, but just 10% of their total spending. Table 1: Catastrophic impact of out-of-pocket payments, Vietnam, 1993, various thresholds threshold level z Out-of-pocket health expenditure 5% 10% 15% 25% as % of total expenditure per cap Headcount (H) 38.19% 18.40% 9.26% - Mean gap (G) 2.85% 1.51% 0.84% - Mean positive gap (MPG) 7.47% 8.21% 9.06% - as % of non-food expenditure per cap Headcount (H) 67.17% 46.52% 33.25% 17.88% Mean gap (G) 9.95% 7.14% 5.17% 2.70% Mean positive gap (MPG) 14.81% 15.36% 15.55% 15.11% Figure 2: Catastrophic payments as share of total and non-food expenditure 1 oop as % of spending % 20% 40% 60% 80% 100% cum% of pop, ranked by decreasing oop share oop as % of tot exp oop as % of non-food exp Catastrophic payments Page 3

4 Making catastrophic impact sensitive to income rank The approach thus far is insensitive to whether it is the poor or the better-off individuals who exceed the threshold. It seems likely that most societies will care more if it is an individual in the lowest decile whose spending (as a share of its income) exceeds the threshold than if it is one in the top decile. One way of registering the location of the incidence and intensity of the impact across the income distribution is to compute a concentration index for i and O i, which we could define as C and C O. [See Technical Notes 5 and 6 for the definition of a concentration curve and index]. A positive value of C indicates a greater tendency for the better-off to exceed the payment threshold, while a negative value indicates that the worse-off are more likely to exceed the threshold. Similarly, a positive value of C O indicates a greater intensity of overshoots amongst the better-off, whilst a negative value will indicate a greater intensity amongst the worse-off. But because the level measures (i.e. Η and G) and the distribution measures (i.e. C and C O )ofthe catastrophic impact could vary in the same or in opposite directions, it is useful to have an index trading off the two dimensions. This can be achieved by constructing weighted versions of the level measures which take into account whether the excesses are concentrated mostly amongst the poor or the better-off. One way of doing this is by weighting the excess indicator variable, i, by the individual's rank in the income distribution. One particular weighting scheme (proposed in [1]) attaches to the person with the lowest standard of living a weight equal to 2, and then the weight declines linearly with rank so that the person with the highest level of living receives a weight of zero. This leads to a (rank-dependent) weighted index W defined as [1]: (5) W H ( 1 C ) =. The weighting results in an attractive and simple summary measure defined as the catastrophic payment headcount multiplied by the complement of the concentration index. If those who exceed the threshold tend to be poorer, the concentration index C will be negative, and this will make W greater than Η. Thus the catastrophic payment problem is worse than it appears simply by looking at the fraction of the population exceeding the threshold, since it overlooks the fact that it tends to be the poor who exceed the threshold. By contrast, if it is the better-off individuals who tend to exceed the threshold, C will be positive, and Η will overstate the problem of the catastrophic payments as measured by W. The same weighting approach can be used for the catastrophic payment excess to define: G (6) W G ( 1 C O ) =. As above, the difference between the unweighted G and the rank-weighted W G depends on the G distribution of the overshoots: W will be larger than G to the extent that a greater share of the overshoots occurs amongst the poorer population groups (i.e. if C is negative). O Box 2: Rank-sensitive measures of catastrophic impact Table 2 presents the concentration indices and the rank-weighted headcount and gap measures for the same example of Vietnam. Again, using a different denominator makes a substantial difference. Not only looks the catastrophic impact much larger when oop payments are expressed as a fraction of discretionary expenditure, also the distribution by income is quite different. For shares of total expenditure, the concentration indices of headcounts and gaps are are mostly positive, indicating greater incidence and Catastrophic payments Page 4

5 intensity among the better-off. As a result, the rank-weighted measures are smaller than the unweighted measures in Table 1. The opposite is true for the shares of discretionary expenditure: (generally) negative concentration indices suggest catastrophic impact is more prevalent and more intense among poorer Vietnamese households, and the rank-weighted measures therefore tend to be larger than the unweighted ones. Table 2: Rank-weighted incidence (headcount) and intensity (or gap) of catastrophic out-of-pocket payments in Vietnam, 1993 threshold level z Out-of-pocket health expenditure 5% 10% 15% 25% as % of total expenditure per cap Conc index of incidence (C ) Rank-weighted headcount ( W ) 38.57% 18.46% 9.30% - Conc index of intensity (C O ) O Rank-weighted headcount ( W ) 2.85% 1.51% 0.84% - as % of non-food expenditure per cap Conc index of incidence (C ) Rank-weighted headcount ( W ) 71.50% 50.40% 35.75% 18.98% Conc index of intensity (C O ) O Rank-weighted headcount ( W ) 10.49% 7.46% 5.32% 2.68% Computing catastrophic impact measures from micro data When micro-level data on household payments and living standards are available, the measures outlined above can be computed fairly straightforwardly using routine statistical packages such as SPSS or Stata. SPSS syntax The first step is to generate a fraction variable indicating health care payments as a proportion of the living standards (or ability to pay) variable. In the example below, based on data from the 1993 Vietnam Living Standards Survey (VLSS), this variable oopshare is computed as per capita out-of-pocket health payments, pcoop, divided by per capita household consumption, pcexp. This is the variable for the y- axis in the payment fraction chart. [Remember to use the WIGHT commandifthesampleisnotself- weighting] compute oopshre = pcoop/pcexp. compute ceiling = 0.2. Catastrophic payments Page 5

6 [Obviously, the procedure is identical if instead of pcexp a variable like non-food expenditure (e.g. pcnfexp) is used.] We then identify those facing catastrophic expenditure are identified by catcount and the extent to which this is the case by catgap, and compute their means. compute catcount = 0. compute catgap = 0. if (oopshre > ceiling) catcount = 1. If (oopshre > ceiling) catgap = oopshre - ceiling. Descriptives variables= catcount, catgap, oopshre /statistics = mean stddev min max. execute. The means of catcount and catgap directly give us estimates of H and G, while their covariances with the fractional rank can be used to generate the concentration indices using the convenient covariance or regression approach given in TN#7. Covariances can be generated with the usual commands outlined in Technical Note #7. rank variables=pcexp (A) /rank /rfraction /print=yes /ties=mean. rename variables rfr001 = frpcexp. correlations /variables=catcount catgap frpcexp /statistics xprod/ missing = pairwise. execute. Probably the best option to generate the graphs is to copy the oopshre variable(s) into an xcel spreadsheet, to sort in descending order of oopshre, to generate a new rank variable (rfoopsh) andthen do a scatterplot of oopshre versus rfoopsh. The result can be seen in Figure 2. Alternatively, such graphs can also be done in SPSS using the following graph command. rank variables=oopshre (D) /rank /rfraction /print=yes /ties=mean. rename variables rfr001 = rfoopsh. Variable labels oopshre out-of-pocket exp as % of income variable labels rfoopsh "cum prop of oop share". graph /scatterplot(bivar)= rfoopsh with oopshre /missing = listwise. Double clicking on the resultant graph will allow it to be edited. On the axis options, title the axes, set the ranges, check the grid box and edit the legends. The smallest possible marker needs to be selected. Some of the options can be saved in a chart template. Fig 3 shows the SPSS chart obtained with the 1993 Vietnam data. Catastrophic payments Page 6

7 Fig3: Out-of-pocket payment share curve, Vietnam 1993 (SPSS) out-of-pocket exp as % of income cum prop of oop share Stata syntax First we need to generate the relevant catcount and catgap variables. gen zcat =.10 gen oopshre = pcoop / pcexp gen catgap = oopshre - zcat replace catgap = 0 if catgap < 0 gen catcount = (catgap>0) [Obviously, the procedure is identical if instead of pcexp a variable like non-food expenditure (e.g. pcnfexp) is used.] Then we can compute the concentration indices for both measures using the familiar (cf Techn Note #7) methods. I.e. use the glcurve command to generate the fractional rank of the pcexp variable and then obtain the relevant means and covariances for the convenient covariance formula of a concentration index. [Remember to add [fw=wt] in these commands if data are weighted.] glcurve pcexp, pvar (exprank) cor exprank catcount catgap, c m All headcount and gap measures can then be obtained from these results. Alternatively, the relevant variables can be copied into an xcel spreadsheet for the generation of charts with more chart editing options Again, graphs can be produced in xcel by copying the relevant variables The glcurve and graph commands allow an oop share curve diagram to be constructed as shown in Fig 3. * generate the fractional rank variable p in descending order of oopshre gen recshre = 1 - oopshre glcurve oopshre, gl(glshre) pvar(p) sortvar(recshre) * graph the sharecurve gr oopshre p, border psize(10) xlab(0(0.1)1) l1 ("cumul pop prop") l2 ("oops as shre of expend") Catastrophic payments Page 7

8 When copying and pasting the graph, make sure to save it with a white background. This is done by selecting General preferences, Result Colors, Color Scheme: white background to obtain the result in Figure 4. Fig 4: Out-of-pocket payment share curves, Vietnam 1993 (Stata) cumul pop prop oops as shre of atp Cum. Pop. Prop. Useful links Technical Note #7 on the concentration index in this series provides full computational details for weighted and unweighted samples. Bibliography 1. Wagstaff, A. and. van Doorslaer, Catastrophe and Impoverishment in Paying for Health Care: With Applications to Vietnam Health conomics 2003 (forthcoming). 2. Gertler, P. and J. Gruber, Insuring consumption against illness, American conomic Review, 2002, Catastrophic payments Page 8