Public Health (1991), 105, 121-126
© The Society of Public Health, 199l
Predicting Costs and Outcomes of Neonatal Intensive Care for Very Low Birthweight Infants R. C. S t e v e n s o n ] P. O. D. Pharoah, 2 R. W. I. Cooke 3 and B. S a n d h u 1
1Department of Economics, 2Department o f Pub/ic Hea/th and 3Department of Chi/d Hea/th, Liverpoo/ UniversitV
A geographically determined cohort of all infants of less than 1,500 g born in 1980 and 1981 were clinically followed up to age four to determine their disabilities which were classified as mild, moderate or severe. A quality adjustment coefficient, determined by the severity of the disability, was used to calculate the number of quality adjusted lives produced. The total cost of care for these children was assessed as the sum of hospital costs to age four (which were specifically determined) and an estimate of the life-time costs of care of disabled children from information provided by the Education and Social Service departments. A very poor predictive power of birthweigfit with cost was obtained with the ungrouped birthweight data, whereas clinical factors explained up to 60% of the variance of the initial hospital costs for survivors and up to 30% of the variance of lifetime costs and the cost of quality adjusted lives produced. Rules for the allocation of resources based on discrimination by birthweight are flawed but the application of clinical discretion is important.
Introduction N e o n a t a l intensive care ( N I C ) has d e v e l o p e d r a p i d l y d u r i n g the p a s t t w o d e c a d e s a n d has a b s o r b e d a n i n c r e a s i n g a m o u n t o f h e a l t h service resources. It has been a s s o c i a t e d with a r a p i d decline in m o r t a l i t y b u t an increased p r e v a l e n c e o f physical, m e n t a l o r s e n s o r y i m p a i r m e n t . 1'~ T h i s has called i n t o q u e s t i o n the b a l a n c e b e t w e e n c o s t s a n d benefits o f N I C for very low b i r t h w e i g h t ( V L B W ) infants. In an influential p a p e r f r o m C a n a d a , the V L B W r a n g e was stratified into t w o classes; 500-999 g a n d 1,000-1,499 g.3 By a n y o f several m e a s u r e s o f o u t c o m e , infants in the h i g h e r r a n g e benefitted m o r e f r o m N I C t h a n d i d the smallest infants. A c o m p a r i s o n o f costs a n d benefits for i n f a n t s w e i g h i n g less t h a n 1,000 g at b i r t h s h o w e d a net e c o n o m i c loss a t a n y positive d i s c o u n t rate. O n this evidence, a t e n t a t i v e rule was s u g g e s t e d that, w h e r e resources are i n a d e q u a t e to treat all infants, they m i g h t b e t t e r be c o n c e n t r a t e d in the h i g h e r b i r t h w e i g h t ( B W ) range. In an earlier p a p e r which r e p o r t e d on a d e t a i l e d c o s t i n g o f the M e r s e y R e g i o n N I C U n i t , we suggested t h a t B W was p r o b a b l y n o t a g o o d g u i d e to o u t c o m e , b u t at t h a t time we were a b l e o n l y to t a k e a c c o u n t o f costs i n c u r r e d in the n e o n a t a l period. 4 U s i n g new d a t a , we c a n n o w test the usefulness o f the B W rule for p r e d i c t i n g l o n g - t e r m costs a n d o u t c o m e s .
Correspondence: R.C. Stevenson, Department of Economics, Liverpool University, P.O. Box 147, Liverpool, L69 3BX.
R.C. Stevenson et al.
122
Methods: Outcomes The geographically determined sample comprised all infants of BW less than 1,500 g born in 1980 and 1981 to mothers resident in two Mersey Health Districts. It was a sub-set of a larger population which has been reported previously.: There were 109 V L B W infants, all of w h o m were treated in the Regional N I C Unit. Mortality was 37%, and the 69 survivors were followed up and assessed at the age o f four years. Disabilities were classified on a 4 point scale as follows: mild (class 2), moderate (class 3) and severe (class 4). N o r m a l children were allocated to category 1. Expected life span was assumed to be 70 years for all except the severely a b n o r m a l children in category 4 for w h o m it was reduced to 45 years. Disabilities were detected in 11 (16%) of the children. Three were classed as mild; one was moderate and seven were severe. Table I shows the number o f infants, survivors and their disabilities in 100 g BW ranges. The final column of Table I gives the n u m b e r o f quality adjusted lives (QALs) produced. This was calculated by allocating a quality adjustment coefficient to disabled children according to the severity of their abnormality. In the absence of any systematic judgments specific to this study, the coefficients were arbitrarily distributed at equal intervals on a zero to one scale. Mild, moderate and severely disabled children's lives were taken as 0.75, 0.50 and 0.25 of a normal life. It is possible to argue that the 0.25 coefficient attached to the severely a b n o r m a l category might be too high for some infants who would be placed close to the b o t t o m of m o s t scales, including the one used in the Canadian study which allowed a negative value to be placed on some lives) In a sample with a larger n u m b e r o f impaired infants, it would be important to establish a less arbitrary system for assessing the quality o f life, but in this study the relationship between BW and outcome was shown to be insensitive to the selection of the quality adjustment coefficients. Results were also computed in terms of quality adjusted life years (QALYs) but are not presented since they do not differ substantially from those expressed in QALs. Costs Total costs in Table II were estimated as the sum of neonatal costs from birth to first discharge home, postnatal costs to the age o f four years and projected life-time costs for abnormal children. They do not represent the full cost to society since, for instance, no Table I Birthweight (g) 500-599 600-699 700-799 800-899 900-999 1000-1099 1100-1199 1206-1299 1306-1399 1400-1499
Medical outcomes for 109 VLBW infants
Infants
Survivors
2 5 9 12 9 |7 16 10 20 9
0 2 1 6 3 11 10 10 18 8
Disabilities class 2 3 4 0 0 0 1 0 1 0 1 0 0
0 0 0 0 0 0 1 0 0 0
0 0 1 0 I 1 0 2 2 0
Number of QALs 0 2.00 0.25 5.75 2.25 10.00 9.50 8.25 16.50 8.00
V L B Infants - Costs and O u t c o m e s o f Care
Table II
123
Neonatal, postnatal and life-time costs of care for 109 VLBW infants 1984 pay and prices. (Discounted at 5%)
Birthwcight (g)
Total cost (£)
Average cost of a survivor (£)
Average cost of a QAL (£)
500-599 600-699 700-799 800-899 900-999 1000-1099 1100-1199 1200-1299 1300-1399 1400-1499
1,188 33,000 146,012 140,267 114,187 235,836 210,472 276,746 275,011 61,890
16,500 146,012 23,378 38,062 21,440 21,047 27,675 15,278 7,736
16,500 584,048 24,394 50,750 23,584 22,155 33,545 16,667 7,736
attempt was made to measure private pecuniary and non-pecuniary costs to families. Nevertheless, the estimates capture a high proportion o f exchequer costs. Initial hospital costs included in-patient days in three Liverpool hospitals before or after treatment in the Regional N I C Unit. The cost o f N I C was derived from an earlier study which was a full costing o f the Unit. 4 It involved the estimation o f day costs for three levels of care: intensive, special and nursery. Each infant was costed individually according to the number o f days spent in each care level. Surgery and diagnostic tests were costed separately. Health service costs after the first discharge to h o m e were made up of in-patient day and out-patient visits for each child up to the age of four. W o r k in progress will re-assess each child at the age of eight. It will identify and evaluate delays in educational development and subtler sorts o f disability which may not have been evident at earlier ages. F o r the purposes o f the present paper, it was assumed the V L B W children assessed as normal at age four, would not incur long-term medical costs above those of a child of normal BW. An estimate was made o f the life-time costs o f care for disabled children from information provided by the Liverpool Education and Social Services Department. Moderately and severely abnormal children were assumed to require special nursery education when aged three and four, at a cost of £6,000 per child; special education from 5-15 years at £4,675 a year and from 16--18 years at £3,430 a year and institutional care from 19 to death, at £7,904 a year. Some of these disabled children and adults would receive part of their care at home but it was assumed that the cost to their families would be similar to the exchequer cost o f institutional care. Mildly abnormal children were assumed to require no special care. In Table II costs are expressed in 1984 pay and prices. Since these costs occur at different times, they are discounted at 5%. Column 2, in Table II, gives the cost per survivor. It is the total cost of caring for survivors and non-survivors divided by the number o f survivors. Column 3 adjusts for the quality of life and gives the cost of producing a QAL. It is total cost divided by the number of QALs produced in each BW range. Relationship between BW and Outcome
The usefulness of the rule proposed in the Canadian study for allocating N I C depends on the existence o f a strong, stable relationship between BW and outcome where outcome
124
R . C . Stevenson et al.
takes account o f mortality, impairment and the long-term costs o f care. The Liverpool data, like the Canadian, gave some credence to this proposition provided that it was presented in b r o a d classes. In the 1980-81 sample, an average Q A L cost £42,405 for infants o f BW less than 1,000 g; this c o m p a r e d with £20,286 for infants in the 1,000-1,499 g range. On average, infants in the lower range were more than twice as expensive as those in the higher range and further evidence was found when the data was analysed in 100 g BW class intervals. Ordinary least squares regression (OLS) o f BW (mid-range points) against the cost of producing a survivor (which included the costs of non-survivors) gave: Cost--= £231,845 - 171.6 BW n - 9 ; r2=0.57; P < 0 . 0 1 The same data, adjusted for the quality of life, gave a weaker relationship. In a much less robust equation, r 2= 0.23 (not significant). The first equation provided some support for the BW hypothesis, although over 40% o f the variation in cost remained unexplained; but the weakness of grouping the birthweight data was demonstrated by arranging the same data in different 100 g ranges (450-549 g, 550-649 g etc., instead of 500-599 g, 600-699 g etc.). OLS regression of BW on the cost of producing a survivor now gave: Cost = £ 5 7 , 9 0 2 - 30.55 BW n-- 10; r2-=0.10; not significant In a similar equation relating BW to the cost of producing a QAL, r was not significantly different from zero. Similar tests were performed on the whole sample of 109 infants. The very p o o r predictive power o f BW over costs was confirmed for both survivors and non-survivors. Most significantly, there was no sign of a structural break in the data at either 1000 g or 900 g. Casual inspection suggested that, if any such break did exist, it was at a b o u t 800 g but the small number o f observations in the lowest BW ranges made it impossible to m a k e a statistically reliable inference. The failure to find a strong, simple and stable relation between BW and cost suggested that, data problems aside, the model was incomplete. Therefore the model was expanded to include a variety o f pre-clinical and clinical variables which might explain more of the variance in costs and outcomes. Pre-clinical and Clinical Factors Statistical regularities were sought between various measures o f cost and outcome and: BW 2= a simple test for non-linearity. S = a d u m m y variable; survivor = 1, non-survivor = 0. B ---a d u m m y variable; inborn = 1, outborn = 0. I U T = a d u m m y variable; in utero transfer = 1, others = 0. G = gestational age in weeks. A large number o f equations were tested. Most coefficients and equations were significant at P < 0.05 but their explanatory power was weak and frequently not significantly different from zero. Neither BW nor BW 2 added much to the closeness of fit and the explanatory power of the models was not improved by restricting the sample to survivors, non-survivors, in-born or those infants who were ventilated for at least one day.
V L B Infants - Costs and Outcomes o f Care
125
To test the influence of clinical factors on cost and outcome, four conditions were defined which were expected to add to the intensity and duration o f care. These were respiratory distress syndrome, septicaemia, intraventricular haemorrhage and necrotising enterocolitis. They were represented by d u m m y variables in a variety of step-wise and OLS regression models. In equations significant at P < 0.05, clinical factors explained up to 60% o f the variance in initial admission hospital costs for survivors. They were not significant in explaining N I C costs for non-survivors. As was expected, they were less powerful in explaining life-time costs and the cost of a QAL, but in equations significant at P < 0.05, they still explained up to 30% of the variance in cost. No significant difference could be detected in the effect o f clinical factors on the cost o f care for infants o f BW greater or less than 1000 g. Discussion Clinicians and economists have analysed the effectiveness of N I C in terms of its differential impact on infants grouped according to BW specific class intervals. N o justification for the selection o f these class intervals has been offered in terms o f clinical intuition or structural breaks in the data. Where full information is available, grouping by BW range is statistically inefficient and m a y be misleading when important differences exist between infants within BW ranges. In a larger model which used ungrouped data, no systematic relationship was detected between BW and outcomes. BW did have some explanatory power over grouped data but this was shown to depend on the selection o f BW class intervals. There was a correlation between mortality and BW but good and relatively inexpensive outcomes were obtained within the lowest ranges and some poor, expensive outcomes were found amongst the larger infants. Attempts to predict the incidence o f disability from BW, clinical and preclinical factors were unsuccessful. It was not possible, perhaps because the numbers were too few, to reject the hypothesis that abnormalities in V L B W infants are a r a n d o m event. This being the case, it was not surprising that BW could not predict the cost o f a Q A L since life-time costs were dominated by the high cost of caring for a b n o r m a l children. When these large life-time costs were omitted, BW still did not predict costs and outcomes even in the neonatal period or the period up to the age of four. The findings have been confirmed in a similar analysis o f the 1979 and 1983 cohorts treated at the Mersey N I C Unit. Attempts to increase the explanatory power of the model by taking account of preclinical and clinical factors were only partially successful. Preclinical variables were found to have little influence on costs. Clinical factors were shown to be much more important in at least some sub-sets o f the data. Although no evidence was found of a structural break at 1,000 g or 900 g, caution has to be exercised in interpreting the Mersey Region N I C Unit. Small numbers o f observations in the lowest BW ranges m a y have biased the findings. A single severely disabled survivor gave the cost of Q A L as £584,048 in the 700-799 g range. There was a temptation to exclude this patient as a statistical 'outlier' but each cohort which we have examined has included at least one case of this sort. I f these few expensive infants with p o o r outcomes are not taken into account, an important aspect of the uncertain nature of N I C would be missed. Furthermore, this one biases the results against our general findings. The interest in predicting costs and outcomes for N I C derives from an urge to devise rational criteria for the allocation of care where resources are inadequate to permit the treatment of all patients. Rational criteria embodied in rules may be regarded as a guide for
126
R . C . Stevenson et al.
planning future services and an aid to clinicians in the formulation o f policy in existing N I C units. Probably most clinicians would accept the validity o f this approach, at least for planning purposes, but rules have to be good rules. As a minimum requirement, rules should make meaningful distinctions between classes o f patients who are to be treated and those who are not. In a complex therapy such as NIC, it is unlikely that a single characteristic such as BW would be adequate. Special care is needed in defining the boundaries between classes of patients and the rule should be reasonably stable over time and between hospitals. This is unlikely to be achieved in a changing technological environment where wide variations are known to exist in clinical practice and cost efficiency. Above all rules must be clinically acceptable; at least they should make provision for the application of clinical discretion. On the evidence of one Regional N I C Unit, rules for the allocation of care based on discrimination by BW do not seem to possess any of these characteristics.
Acknowledgements This work was supported by grants from the Mersey Region Research Committee, The University of Liverpool and the DHSS. We are also indebted to the staffof the Mersey Region NIC Unit and to Dr L. Rosenbloom, consultant in paediatric neurology at the Royal Liverpool Childrens' Hospital.
References 1. Stewart, A. L., Reynolds, E. O. R. & Lipscombe, A. P. (1981). Outcome for infants of very low birthweight; survey of world literature. Lancet, i, 1038 1040. 2. Pharoah, P. O. D., Stevenson, R. C., Cooke, R. W. I. & Sandhu, B. (1988). The costs and benefits of neonatal intensive care. Archives of Diseases in Childhood, 63, 715 718. 3. Boyle, M. H., Torrance, G. W., Sinclair, J. C. & Horwood, S. P. (1983). Economic evaluation of neonatal intensive care of very low birthweight infants. New England Journal of Medicine, 308, 1330-1337. 4. Sandhu, B., Stevenson, R.C., Cooke, R. W. I. & Pharoah, P. O. D. (1986). Cost of neonatal intensive care for very low birthweight infants. Lancet, i, 600-603.
© The Society of Public Health, 199l
Predicting Costs and Outcomes of Neonatal Intensive Care for Very Low Birthweight Infants R. C. S t e v e n s o n ] P. O. D. Pharoah, 2 R. W. I. Cooke 3 and B. S a n d h u 1
1Department of Economics, 2Department o f Pub/ic Hea/th and 3Department of Chi/d Hea/th, Liverpoo/ UniversitV
A geographically determined cohort of all infants of less than 1,500 g born in 1980 and 1981 were clinically followed up to age four to determine their disabilities which were classified as mild, moderate or severe. A quality adjustment coefficient, determined by the severity of the disability, was used to calculate the number of quality adjusted lives produced. The total cost of care for these children was assessed as the sum of hospital costs to age four (which were specifically determined) and an estimate of the life-time costs of care of disabled children from information provided by the Education and Social Service departments. A very poor predictive power of birthweigfit with cost was obtained with the ungrouped birthweight data, whereas clinical factors explained up to 60% of the variance of the initial hospital costs for survivors and up to 30% of the variance of lifetime costs and the cost of quality adjusted lives produced. Rules for the allocation of resources based on discrimination by birthweight are flawed but the application of clinical discretion is important.
Introduction N e o n a t a l intensive care ( N I C ) has d e v e l o p e d r a p i d l y d u r i n g the p a s t t w o d e c a d e s a n d has a b s o r b e d a n i n c r e a s i n g a m o u n t o f h e a l t h service resources. It has been a s s o c i a t e d with a r a p i d decline in m o r t a l i t y b u t an increased p r e v a l e n c e o f physical, m e n t a l o r s e n s o r y i m p a i r m e n t . 1'~ T h i s has called i n t o q u e s t i o n the b a l a n c e b e t w e e n c o s t s a n d benefits o f N I C for very low b i r t h w e i g h t ( V L B W ) infants. In an influential p a p e r f r o m C a n a d a , the V L B W r a n g e was stratified into t w o classes; 500-999 g a n d 1,000-1,499 g.3 By a n y o f several m e a s u r e s o f o u t c o m e , infants in the h i g h e r r a n g e benefitted m o r e f r o m N I C t h a n d i d the smallest infants. A c o m p a r i s o n o f costs a n d benefits for i n f a n t s w e i g h i n g less t h a n 1,000 g at b i r t h s h o w e d a net e c o n o m i c loss a t a n y positive d i s c o u n t rate. O n this evidence, a t e n t a t i v e rule was s u g g e s t e d that, w h e r e resources are i n a d e q u a t e to treat all infants, they m i g h t b e t t e r be c o n c e n t r a t e d in the h i g h e r b i r t h w e i g h t ( B W ) range. In an earlier p a p e r which r e p o r t e d on a d e t a i l e d c o s t i n g o f the M e r s e y R e g i o n N I C U n i t , we suggested t h a t B W was p r o b a b l y n o t a g o o d g u i d e to o u t c o m e , b u t at t h a t time we were a b l e o n l y to t a k e a c c o u n t o f costs i n c u r r e d in the n e o n a t a l period. 4 U s i n g new d a t a , we c a n n o w test the usefulness o f the B W rule for p r e d i c t i n g l o n g - t e r m costs a n d o u t c o m e s .
Correspondence: R.C. Stevenson, Department of Economics, Liverpool University, P.O. Box 147, Liverpool, L69 3BX.
R.C. Stevenson et al.
122
Methods: Outcomes The geographically determined sample comprised all infants of BW less than 1,500 g born in 1980 and 1981 to mothers resident in two Mersey Health Districts. It was a sub-set of a larger population which has been reported previously.: There were 109 V L B W infants, all of w h o m were treated in the Regional N I C Unit. Mortality was 37%, and the 69 survivors were followed up and assessed at the age o f four years. Disabilities were classified on a 4 point scale as follows: mild (class 2), moderate (class 3) and severe (class 4). N o r m a l children were allocated to category 1. Expected life span was assumed to be 70 years for all except the severely a b n o r m a l children in category 4 for w h o m it was reduced to 45 years. Disabilities were detected in 11 (16%) of the children. Three were classed as mild; one was moderate and seven were severe. Table I shows the number o f infants, survivors and their disabilities in 100 g BW ranges. The final column of Table I gives the n u m b e r o f quality adjusted lives (QALs) produced. This was calculated by allocating a quality adjustment coefficient to disabled children according to the severity of their abnormality. In the absence of any systematic judgments specific to this study, the coefficients were arbitrarily distributed at equal intervals on a zero to one scale. Mild, moderate and severely disabled children's lives were taken as 0.75, 0.50 and 0.25 of a normal life. It is possible to argue that the 0.25 coefficient attached to the severely a b n o r m a l category might be too high for some infants who would be placed close to the b o t t o m of m o s t scales, including the one used in the Canadian study which allowed a negative value to be placed on some lives) In a sample with a larger n u m b e r o f impaired infants, it would be important to establish a less arbitrary system for assessing the quality o f life, but in this study the relationship between BW and outcome was shown to be insensitive to the selection of the quality adjustment coefficients. Results were also computed in terms of quality adjusted life years (QALYs) but are not presented since they do not differ substantially from those expressed in QALs. Costs Total costs in Table II were estimated as the sum of neonatal costs from birth to first discharge home, postnatal costs to the age o f four years and projected life-time costs for abnormal children. They do not represent the full cost to society since, for instance, no Table I Birthweight (g) 500-599 600-699 700-799 800-899 900-999 1000-1099 1100-1199 1206-1299 1306-1399 1400-1499
Medical outcomes for 109 VLBW infants
Infants
Survivors
2 5 9 12 9 |7 16 10 20 9
0 2 1 6 3 11 10 10 18 8
Disabilities class 2 3 4 0 0 0 1 0 1 0 1 0 0
0 0 0 0 0 0 1 0 0 0
0 0 1 0 I 1 0 2 2 0
Number of QALs 0 2.00 0.25 5.75 2.25 10.00 9.50 8.25 16.50 8.00
V L B Infants - Costs and O u t c o m e s o f Care
Table II
123
Neonatal, postnatal and life-time costs of care for 109 VLBW infants 1984 pay and prices. (Discounted at 5%)
Birthwcight (g)
Total cost (£)
Average cost of a survivor (£)
Average cost of a QAL (£)
500-599 600-699 700-799 800-899 900-999 1000-1099 1100-1199 1200-1299 1300-1399 1400-1499
1,188 33,000 146,012 140,267 114,187 235,836 210,472 276,746 275,011 61,890
16,500 146,012 23,378 38,062 21,440 21,047 27,675 15,278 7,736
16,500 584,048 24,394 50,750 23,584 22,155 33,545 16,667 7,736
attempt was made to measure private pecuniary and non-pecuniary costs to families. Nevertheless, the estimates capture a high proportion o f exchequer costs. Initial hospital costs included in-patient days in three Liverpool hospitals before or after treatment in the Regional N I C Unit. The cost o f N I C was derived from an earlier study which was a full costing o f the Unit. 4 It involved the estimation o f day costs for three levels of care: intensive, special and nursery. Each infant was costed individually according to the number o f days spent in each care level. Surgery and diagnostic tests were costed separately. Health service costs after the first discharge to h o m e were made up of in-patient day and out-patient visits for each child up to the age of four. W o r k in progress will re-assess each child at the age of eight. It will identify and evaluate delays in educational development and subtler sorts o f disability which may not have been evident at earlier ages. F o r the purposes o f the present paper, it was assumed the V L B W children assessed as normal at age four, would not incur long-term medical costs above those of a child of normal BW. An estimate was made o f the life-time costs o f care for disabled children from information provided by the Liverpool Education and Social Services Department. Moderately and severely abnormal children were assumed to require special nursery education when aged three and four, at a cost of £6,000 per child; special education from 5-15 years at £4,675 a year and from 16--18 years at £3,430 a year and institutional care from 19 to death, at £7,904 a year. Some of these disabled children and adults would receive part of their care at home but it was assumed that the cost to their families would be similar to the exchequer cost o f institutional care. Mildly abnormal children were assumed to require no special care. In Table II costs are expressed in 1984 pay and prices. Since these costs occur at different times, they are discounted at 5%. Column 2, in Table II, gives the cost per survivor. It is the total cost of caring for survivors and non-survivors divided by the number o f survivors. Column 3 adjusts for the quality of life and gives the cost of producing a QAL. It is total cost divided by the number of QALs produced in each BW range. Relationship between BW and Outcome
The usefulness of the rule proposed in the Canadian study for allocating N I C depends on the existence o f a strong, stable relationship between BW and outcome where outcome
124
R . C . Stevenson et al.
takes account o f mortality, impairment and the long-term costs o f care. The Liverpool data, like the Canadian, gave some credence to this proposition provided that it was presented in b r o a d classes. In the 1980-81 sample, an average Q A L cost £42,405 for infants o f BW less than 1,000 g; this c o m p a r e d with £20,286 for infants in the 1,000-1,499 g range. On average, infants in the lower range were more than twice as expensive as those in the higher range and further evidence was found when the data was analysed in 100 g BW class intervals. Ordinary least squares regression (OLS) o f BW (mid-range points) against the cost of producing a survivor (which included the costs of non-survivors) gave: Cost--= £231,845 - 171.6 BW n - 9 ; r2=0.57; P < 0 . 0 1 The same data, adjusted for the quality of life, gave a weaker relationship. In a much less robust equation, r 2= 0.23 (not significant). The first equation provided some support for the BW hypothesis, although over 40% o f the variation in cost remained unexplained; but the weakness of grouping the birthweight data was demonstrated by arranging the same data in different 100 g ranges (450-549 g, 550-649 g etc., instead of 500-599 g, 600-699 g etc.). OLS regression of BW on the cost of producing a survivor now gave: Cost = £ 5 7 , 9 0 2 - 30.55 BW n-- 10; r2-=0.10; not significant In a similar equation relating BW to the cost of producing a QAL, r was not significantly different from zero. Similar tests were performed on the whole sample of 109 infants. The very p o o r predictive power o f BW over costs was confirmed for both survivors and non-survivors. Most significantly, there was no sign of a structural break in the data at either 1000 g or 900 g. Casual inspection suggested that, if any such break did exist, it was at a b o u t 800 g but the small number o f observations in the lowest BW ranges made it impossible to m a k e a statistically reliable inference. The failure to find a strong, simple and stable relation between BW and cost suggested that, data problems aside, the model was incomplete. Therefore the model was expanded to include a variety o f pre-clinical and clinical variables which might explain more of the variance in costs and outcomes. Pre-clinical and Clinical Factors Statistical regularities were sought between various measures o f cost and outcome and: BW 2= a simple test for non-linearity. S = a d u m m y variable; survivor = 1, non-survivor = 0. B ---a d u m m y variable; inborn = 1, outborn = 0. I U T = a d u m m y variable; in utero transfer = 1, others = 0. G = gestational age in weeks. A large number o f equations were tested. Most coefficients and equations were significant at P < 0.05 but their explanatory power was weak and frequently not significantly different from zero. Neither BW nor BW 2 added much to the closeness of fit and the explanatory power of the models was not improved by restricting the sample to survivors, non-survivors, in-born or those infants who were ventilated for at least one day.
V L B Infants - Costs and Outcomes o f Care
125
To test the influence of clinical factors on cost and outcome, four conditions were defined which were expected to add to the intensity and duration o f care. These were respiratory distress syndrome, septicaemia, intraventricular haemorrhage and necrotising enterocolitis. They were represented by d u m m y variables in a variety of step-wise and OLS regression models. In equations significant at P < 0.05, clinical factors explained up to 60% o f the variance in initial admission hospital costs for survivors. They were not significant in explaining N I C costs for non-survivors. As was expected, they were less powerful in explaining life-time costs and the cost of a QAL, but in equations significant at P < 0.05, they still explained up to 30% of the variance in cost. No significant difference could be detected in the effect o f clinical factors on the cost o f care for infants o f BW greater or less than 1000 g. Discussion Clinicians and economists have analysed the effectiveness of N I C in terms of its differential impact on infants grouped according to BW specific class intervals. N o justification for the selection o f these class intervals has been offered in terms o f clinical intuition or structural breaks in the data. Where full information is available, grouping by BW range is statistically inefficient and m a y be misleading when important differences exist between infants within BW ranges. In a larger model which used ungrouped data, no systematic relationship was detected between BW and outcomes. BW did have some explanatory power over grouped data but this was shown to depend on the selection o f BW class intervals. There was a correlation between mortality and BW but good and relatively inexpensive outcomes were obtained within the lowest ranges and some poor, expensive outcomes were found amongst the larger infants. Attempts to predict the incidence o f disability from BW, clinical and preclinical factors were unsuccessful. It was not possible, perhaps because the numbers were too few, to reject the hypothesis that abnormalities in V L B W infants are a r a n d o m event. This being the case, it was not surprising that BW could not predict the cost o f a Q A L since life-time costs were dominated by the high cost of caring for a b n o r m a l children. When these large life-time costs were omitted, BW still did not predict costs and outcomes even in the neonatal period or the period up to the age of four. The findings have been confirmed in a similar analysis o f the 1979 and 1983 cohorts treated at the Mersey N I C Unit. Attempts to increase the explanatory power of the model by taking account of preclinical and clinical factors were only partially successful. Preclinical variables were found to have little influence on costs. Clinical factors were shown to be much more important in at least some sub-sets o f the data. Although no evidence was found of a structural break at 1,000 g or 900 g, caution has to be exercised in interpreting the Mersey Region N I C Unit. Small numbers o f observations in the lowest BW ranges m a y have biased the findings. A single severely disabled survivor gave the cost of Q A L as £584,048 in the 700-799 g range. There was a temptation to exclude this patient as a statistical 'outlier' but each cohort which we have examined has included at least one case of this sort. I f these few expensive infants with p o o r outcomes are not taken into account, an important aspect of the uncertain nature of N I C would be missed. Furthermore, this one biases the results against our general findings. The interest in predicting costs and outcomes for N I C derives from an urge to devise rational criteria for the allocation of care where resources are inadequate to permit the treatment of all patients. Rational criteria embodied in rules may be regarded as a guide for
126
R . C . Stevenson et al.
planning future services and an aid to clinicians in the formulation o f policy in existing N I C units. Probably most clinicians would accept the validity o f this approach, at least for planning purposes, but rules have to be good rules. As a minimum requirement, rules should make meaningful distinctions between classes o f patients who are to be treated and those who are not. In a complex therapy such as NIC, it is unlikely that a single characteristic such as BW would be adequate. Special care is needed in defining the boundaries between classes of patients and the rule should be reasonably stable over time and between hospitals. This is unlikely to be achieved in a changing technological environment where wide variations are known to exist in clinical practice and cost efficiency. Above all rules must be clinically acceptable; at least they should make provision for the application of clinical discretion. On the evidence of one Regional N I C Unit, rules for the allocation of care based on discrimination by BW do not seem to possess any of these characteristics.
Acknowledgements This work was supported by grants from the Mersey Region Research Committee, The University of Liverpool and the DHSS. We are also indebted to the staffof the Mersey Region NIC Unit and to Dr L. Rosenbloom, consultant in paediatric neurology at the Royal Liverpool Childrens' Hospital.
References 1. Stewart, A. L., Reynolds, E. O. R. & Lipscombe, A. P. (1981). Outcome for infants of very low birthweight; survey of world literature. Lancet, i, 1038 1040. 2. Pharoah, P. O. D., Stevenson, R. C., Cooke, R. W. I. & Sandhu, B. (1988). The costs and benefits of neonatal intensive care. Archives of Diseases in Childhood, 63, 715 718. 3. Boyle, M. H., Torrance, G. W., Sinclair, J. C. & Horwood, S. P. (1983). Economic evaluation of neonatal intensive care of very low birthweight infants. New England Journal of Medicine, 308, 1330-1337. 4. Sandhu, B., Stevenson, R.C., Cooke, R. W. I. & Pharoah, P. O. D. (1986). Cost of neonatal intensive care for very low birthweight infants. Lancet, i, 600-603.
