Cost implications of improving the quality of child care using integrated clinical algorithms: Evidence from Northeast Brazil

1. Introduction 

In December 2005, the Child Survival Countdown conference, the first in a series of two-yearly meetings to review and document progress in child survival, declared that progress in several countries was lagging behind the rest of the world [1], [2]. Among the major obstacles identified were the lack of financial resources and capacity for equitable scale up of key health interventions [3], [4]. Although Brazil as a whole has achieved an 80% reduction in under-five mortality rates over the last four decades, 36 per 1000 in 2002 compared with 177 per 1000 in 1960 [5], major inequalities still exist across and within states. Infant mortality rates in the northeast, for example, the poorest region in Brazil, reach levels similar to those in Africa and Asia while those in the richer states are similar to those in Western Europe [6], [7], [8].

While limited financial resources impose important barriers to increased access to essential health interventions in Brazil as elsewhere, in many settings there is the potential to achieve more with the available resources, e.g., by improving efficiency in delivering health services, and by changing the mix of activities undertaken [9], [10]. This may well be true of child health services in Brazil.

The Integrated Management of Childhood Illness (IMCI), a strategy to address the five leading causes of childhood mortality in the developing world, is a possible candidate [11]. Recent evidence from Tanzania has shown that it leads to better quality of care and possibly lower under-five mortality rates at no extra costs [12]. Studies from Uganda and Bangladesh have also shown that health workers trained in IMCI perform better in assessing and managing sick children, and in counselling their caretakers than those not trained, although they have not as yet reported the cost implications of such training [13], [14], [15]. In northeast Brazil, it has also been demonstrated that health workers trained in IMCI provide better quality care than other providers, and this study seeks to fill an important information gap by reporting on the cost implications of introducing and delivering child care based on IMCI training. It does so by estimating the total and additional costs incurred in the implementation of the Brazilian version of IMCI relative to routine care. The study is part of the Multi-Country Evaluation of IMCI, a WHO project undertaken to evaluate the costs and health impact of the IMCI in five countries, Brazil, Tanzania, Uganda, Peru and Bangladesh [16].

2. Methods 

2.1. Study setting and design 

In 1996, the IMCI strategy was identified as a priority for child health by Brazil's Ministry of Health (MOH), with emphasis on first-level care. IMCI implementation was undertaken in the context of the Family Health Programme (FHP), primarily aimed at increasing access to health services by the poorest segment of the Brazilian population. Accordingly, IMCI training was provided mainly to health workers in facilities covered by the FHP, although training coverage has increased at a much slower rate than FHP coverage. Training consisted of a 7–8-day course and one follow-up visit by IMCI supervisors 2 months later. IMCI implementation was particularly strong in the Northeast region, the poorest area in Brazil, from 1997 to 2001. No additional activities to strengthen health systems or to improve family and community practices, the other two components of IMCI, were implemented except in the State of Pernambuco where steps to improve routine supervision were also taken.

The evaluation of the Brazil version of IMCI had a mixed retrospective-prospective design, because IMCI was already implemented in many municipalities at the start of the survey. The study compares municipalities in each of the four states in the northeast considered to have had stable IMCI implementation since 1999, Bahia, Ceará, Paraíba and Pernambuco, with matched municipalities where routine care based on traditional disease-specific case management continued. Only municipalities with population sizes between 5000 and 50,000 inhabitants were considered for inclusion. Municipalities with IMCI were required to have a minimum of 60% of their health workers who managed sick children trained in IMCI in the previous two calendar years. Comparison municipalities were expected to have operational FHP programs but no IMCI-trained health workers. Finally, IMCI and comparison municipalities were matched based on geographic region and population size in order to select eight IMCI and eight comparison municipalities from each state. This was not possible in all states, however, since the number of IMCI municipalities fulfilling the above-described criteria was less than 8. In addition, some of the originally sampled comparison municipalities had to be replaced by a nearby municipality because IMCI-trained workers from other municipalities or states had been employed. See Amaral et al. (2004) for more detail on the sampling rationale and methods [13].

The final sample included eight IMCI and eight comparison municipalities each in the states of Paraíba and Pernambuco, seven IMCI and seven comparison municipalities in Ceará, and five IMCI and seven comparison municipalities in Bahia. To compensate for the different number of municipalities in each state, the number of facilities sampled was fixed at 12 with IMCI and 12 without IMCI in each state. However, the IMCI municipalities in Bahia did not prove to have the required 60% of providers trained in IMCI during the course of the study because of high staff turnover, so it was excluded from the analysis of costs. Of the remaining 46 municipalities, only 44 had complete data for the analysis (22 IMCI and 22 comparison municipalities). Table 1 summarizes the main characteristics of the primary care facilities included in the study, summarized separately for IMCI and comparison municipalities.

Table 1. Characteristics of the municipalities included in the study (2001 unless otherwise stated)

	Characteristics	IMCI municipalities	Comparison municipalities	P value (t-test)
	Average number of FHP facilities (S.D.)	7.39 (3.87)	6.17 (2.92)	0.23
	Average distance between primary facilities and the centre of the municipality (S.D.)	1.89 (0.63)	1.91 (0.71)	0.88
	Percent of primary facilities near the centre of the municipality	29%	25%
	Percent of primary facilities between the centre and outskirt of municipality	50%	58%
	Percent of primary facility at the outskirt of the municipality	21%	14.5%
	Average number of under-five children per municipality (S.D.)	2168 (1409)	1523 (1006)	0.01

FHP: Family Health Programme.

2.3. Multivariate analysis 

Municipalities with IMCI had a significantly higher number of under-five visits than comparison municipalities. Therefore, the IMCI status of a facility and the number of visits are not assumed to be independent in the multivariate analysis. It is possible that more parents might seek care for their children from facilities where IMCI has been implemented, perhaps because of perceived quality improvements associated with IMCI. In this care, the appropriate estimation technique would be to adopt a two stage least squares model. The purpose of the first stage is to test the hypothesis that the number of visits to a facility is affected by the presence of IMCI. This was estimated using the ivreg command in STATA software as follows.

Eq. (1) Determinants of visits by under-five children to primary care facilities

(1)

where Childvisits is the annual number of under-five visits in the ith facility, K1 is the presence or not of IMCI training; K2–N are the other possible contextual factors; and e is the error term. Possible determinants of visits for which data were available were the average municipality income per capita and the population of children under-five per catchment area around health facilities. The latter variable controls for both the size of the under-five population and the number of facilities per municipality.

The second stage was then estimated as in Eq. (2).

Eq. (2) Determinants of cost per under-five visit at primary facilities

(2)

where UCi is unit cost per under-five visit in the ith facility; X1 is the presence or not of IMCI training; X2 is the instrumented number of child visits from step 1; X3–N are the other possible explanatory and confounding variables; and e denotes the error term. The explanatory variables considered were the number of full time equivalent physicians or alternatively the number of visits per physician per day, the location of the health facility within the municipality and dummy variables to represent state-specific characteristics such as frequency of supervision, salary scales, etc. Table 2 provides a description of the variables considered in both steps of the multivariate analysis and the expected direction of impact.

Table 2. Variables considered for the analysis of determinants of cost per under-five visit and their description

	Var name	Description	Expected direction of impact
	IMCI	Dummy variable for availability of IMCI based care (1=IMCI; 0=Comparison)	To be explored
	Centre	Facility situated in the centre of the municipality	Positive. Better access which may lead to more visits
	Periphery	Facility situated between centre and outskirt of municipality (third group is facilities situated at the outskirt of the municipality)	Positive (lower impact than in Distance1)
	Ceará	Observations from State of Ceará	Unobservable differences
	Paraíba	Observations form State of Paraíba (third group is Pernambuco)	Unobservable differences
	U5visits	log(n) of under-five visits at primary facility	Negative. More visits lead to lower costs due to increased efficiency
	Doctors	log(n) of full time equivalent doctors	Positive. More doctors lead to higher staff costs
	Visits per day	log(n) of visits per doctor per day	Negative. More visits per doctor lead to lower costs due to increased efficiency

For estimation of both models, the continuous dependent and independent variables were transformed into natural logarithms which best approximated a normal distribution. In addition, because almost 30% of the data on medicines were missing, bootstrap estimation techniques with 1000 sample replications with replacement were used to reduce the potential bias in the estimated parameters [22], [23]. We also used robust cluster estimation to account for the correlation of errors between observations belonging to the same municipality [24].

Finally, one-way and multi-way sensitivity analysis were used to explore the sensitivity of the results to the main assumptions such as discount rate (varied between 0% and 5%) and the useful life of start-up activities (varied between 5 and 15 years); and the uncertainty around some variables such as the cost per visit at primary facilities. See Adam et al. (2007) for more detail on the data collection methods and analysis [25].

3. Results 

Table 3 shows the different components of the cost per child in a standard municipality with 2000 children. For 2001, the cost per child of caring for under-fives in IMCI municipalities was US$ 95.02 compared to US$ 97.96 in the comparison municipalities. None of the reported differences in costs per child were statistically significant, however (Fig. 1).

Table 3. Components of cost per under five visit at primary facilities, with and without IMCI (2001 US$)

	Category	IMCI municipality			Comparison municipality
		Mean	S.D.	N	Mean	S.D.	N
	Salary	7.42	5.40	30	9.94	7.34	34
	Medicines	1.02	1.75	20	1.21	1.34	19
	Vaccines	0.58	0.25	30	0.76	0.29	34
	Capital	0.39	0.32	30	0.55	0.42	34
	Other	0.24	0.32	30	0.34	0.53	34
	Total excluding medicines and vaccines	8.06	5.30	30	10.85	7.36	34
	Total including medicines and vaccines added separatelya	9.08			12.06

No differences between IMCI and comparison municipality means are statistically significant.

a No standard deviations are estimated for total costs because of the smaller sample size that had to be used to estimate the consumption of medicines.

View Large Image Download to PowerPoint

Fig. 1. Components of cost of under-five care per child in a standard (standard municipality with 2000 under-fives) municipality (2001 US$). None of the differences between means are significantly different at the 5% level (t-test).

The most important components of the cost per child are at the municipality, hospital and facility levels, together accounting for 99% of the total cost. At the municipality level, the apparently higher cost per child in municipalities with IMCI would be expected because of the start up and post-implementation running costs of IMCI. At the hospital level, hospital costs appeared to be higher in IMCI districts, linked to the higher number of admissions per child year in municipalities with IMCI relative to the comparison municipalities. However, the differences in costs were not statistically significant, as stated earlier. We return to this in discussion section.

At the primary facility level, costs per child were around 10% lower in municipalities with IMCI, though not statistically significant. Table 3 summarizes the average cost per under-five visit in IMCI and comparison facilities as well as the breakdown of the different components of cost. Although unit costs per input were consistently lower in IMCI than in comparison facilities, differences were also not statistically significant at the 5% level.

Perhaps the most interesting observation is the apparently lower staff costs in facilities with IMCI, given the results of the time and motion study showing that IMCI-trained staff spent longer on average in consultation with under-five children (1.26min longer, after controlling for other determinants) [18]. To explain this finding, we explored the level of capacity utilization of health providers in two ways. First we estimated the average number of consultations per doctor per day, which were 11.55 and 8.05 in IMCI and comparison facilities respectively (p=0.07). This suggests that doctor's time was not fully utilized at the time of the study—if we assume a norm of 25 consultations per day and 15min per consultation. Second, we estimated the proportion of non-productive time out of total time spent in the facility. Non-productive time, defined as time not related to clinical or administrative work, comprised 23% and 29% of total time in IMCI and comparison facilities respectively (p=<0.0001). Total salary costs did not increase in facilities with IMCI because staff had sufficient slack time to adjust their work patterns to cope with the longer time they spent with each child. They could provide higher quality care and spend more time with each child by utilizing some of their down time. No new staff had to be employed. The implications of these findings are considered in the discussion section.

A limitation of the simple comparison of means is that some other important factors differed between facilities with and without IMCI. Most importantly, the number of under-five visits to primary health facilities were significantly higher in municipalities with IMCI (2422 versus 1769 in IMCI and comparison facilities, p=0.008). As explained earlier, it is important to check whether the lower unit costs expected to be associated with more visits per child mask any independent effect of IMCI. But before this can be done it was necessary to check if there is an association between the presence of IMCI and the number of visits.

Table 4 reports the results of the multivariate analysis associated with Eq. (1) above. It suggests that the availability of IMCI trained providers does have an independent effect on the number of under-five visits after controlling for the other possible determinants (p=0.005).

Table 4. Determinants of visits by under-five children to primary health facilities-dependent variable: natural log of visits by under-fives

	Variable	β	Bootstrap S.E.	Z	P>|z|
	IMCI	0.34	0.12	2.84	0.005
	U5 in catchment area	−0.24	0.08	−2.90	0.004
	Centre	0.51	0.17	3.02	0.002
	Periphery	0.39	0.18	2.25	0.02
	Ceará	−0.02	0.22	−0.07	0.94
	Paraíba	0.16	0.15	1.04	0.30
	Constant	8.22	0.52	15.67	<0.0001

Adjusted R2=0.29, Wald χ2=37.27, Prob>χ2<0.0001, N=64. Robust cluster estimation (municipality level-number of clusters=42) and 1000 bootstrap replications. U5 in catchment area: log(n) of number of under-fives in the catchment areas around primary facility; see Table 2 for description of the remaining variables.

These results imply that IMCI and the number of visits cannot be treated as independent variables in Eq. (2) above. Accordingly, we explored the impact of IMCI on the cost per visit controlling for other determinants using two-stage least squares techniques. A full description of the variables explored in the model is found in Table 2. The results are reported in Table 5.

Table 5. Two stage estimationa, b of the determinants of cost per under-five visit-dependent variable: natural log of cost per under-five visit

	Variable	β	Bootstrap S.E.	Z	P>|z|
	IMCI	0.04	0.15	0.24	0.81
	U5visits	−0.90	0.24	−3.67	<0.0001
	Ceará	0.51	0.14	3.56	<0.0001
	Paraíba	0.43	0.13	3.23	0.001
	Constant	7.93	2.53	4.88	<0.0001

Adjusted R2=0.48, Wald χ2=32.30, Prob>χ2<0.0001, N=64.

a Instrumented: U5visits (natural log of visits by under-five children); Instruments: IMCI, U5 in catchment area, distance1, distance2, State1, State2. b Robust cluster estimation at municipality level (number of clusters=42) and 1000 bootstrap replications.

It is important to recognize that this equation excludes the cost of medicines from the analysis. Because of the missing data on medicine utilization described earlier, we faced a difficult choice. We could limit the analysis to the 39 facilities with full data, or undertake the analysis for the 64 facilities, but excluding medicine costs. The former option was explored, and the conclusions were the same as using the full sample excluding cost of medicines. Accordingly, we report the second option here.

The results show that cost per visit is not significantly associated with the presence of IMCI training once patient load was included. Each one percent increase in the number of child visits was associated with 0.9% reduction in the cost per visit, all else held constant (p<0.0001). This suggests that IMCI had no independent effect on the non-medicine cost per under-five visit. It did influence costs indirectly, however, through the apparent increase in the overall number of child visits, something discussed in the next section. The combination of lower cost per visit and higher visits per child in IMCI facilities at the time of our study explains the non-significant difference in the overall cost per child at primary facilities between IMCI and comparison municipalities.

Finally, the difference in cost per child between intervention and comparison districts was not sensitive to variation in parameters during sensitivity analysis. The overall conclusion is that we did not find any evidence that IMCI cost more than routine care in northeast Brazil, even after controlling for possible confounding factors.

4. Discussion 

Up-to-date information on the relative costs and health effects of alternative strategies to reduce child mortality, and the expected variability in costs as interventions are scaled up, will ensure that policy makers have the necessary information to judge if they are moving as rapidly as possible towards their objectives. The aim of this study was to estimate the costs of IMCI as it was implemented in Brazil, already known to improve quality of care for under-five children, to determine if implementing it is a good use of scarce health resources.

We found no evidence that treating children using IMCI versus routine care was associated with higher costs at the time of the study. This finding is consistent with findings from Tanzania, the other MCE site where results on the costs of IMCI are currently available [27]. Together with strong evidence on better health worker's performance when they are trained in IMCI [13], this suggests that implementing IMCI in Brazil was good value for money, especially in the northeast where under-five mortality is still very high compared to other parts of the country.

There are some qualifications that should be taken into account when interpreting these findings, however. First, municipalities differed in ways that could affect the cost of child care (e.g., IMCI municipalities had higher number of under-five children, and more primary facilities and hospitals per municipalities than comparison municipalities). In fact, the decision to implement IMCI in the study municipalities may have been influenced by those factors—for example, it has been suggested that IMCI in Northeast Brazil was more likely to be implemented in municipalities that were larger, more developed economically, and closer to the state capital [2]. While we tried to control for some of these differences in multivariate analysis, it is never possible to be sure that all confounding factors have been taken into account in this type of analysis.

Second, it would have been interesting to explore whether the higher under-five hospital admissions observed in IMCI facilities was directly linked to the presence of IMCI or due to other factors such as differences in the availability of hospital beds—as we did for primary health facilities. This would require information on possible determinants of hospital admissions, for example variables such as case mix or the size of the catchment areas around each hospital. These data were not available. However, we expect the number of admissions per child would be the same or lower in municipalities with IMCI because of more appropriate treatment and better referral patterns, so it seems more likely that the higher admissions are due to factors other than IMCI.

Third, while IMCI was not associated with higher observed costs during the study period, this is only true from a strictly financial point of view and because clinical staff had considerable slack time in our study setting. Providers trained in IMCI spent slightly longer on average (1.26min, p=<0.001) in each consultation with a child under-five than those providing routine care [18]. They also saw more patients per day, as reported earlier. This result might not be generalizable to the rest of Brazil, to other countries, or over time. In situations where providers have little slack time, where absence from work reflects the fact that many health workers hold multiple positions, or where unproductive time is simply a result of inadequate supervision, the additional work load observed in IMCI facilities in our study would not be absorbed so easily and additional staff may have to be employed, thereby increasing the wage bill.

Consideration of competing demands on staff time is also important from an economic, rather than purely financial, perspective, regardless of whether or not down time exists. In theory, the additional time spent in seeing under-fives could have been spent in some other productive activity, something that is particularly relevant when simultaneous efforts to scale up health interventions are in place or in areas already functioning at or near full capacity.

Finally, because of the missing data on medicine utilization described earlier, we faced a difficult choice. We could limit the analysis to the 39 facilities with full data, or undertake the analysis for the 64 facilities, but excluding medicine costs. The former option was explored, and the conclusions were the same as using the full sample excluding cost of medicines. Accordingly, we report the second option here.

In summary, however, this paper makes an important contribution by showing that IMCI, or at least the Brazilian version of it, did not cost more than routine care. This is consistent with the only other study of the costs of implementing IMCI—undertaken in a very different type of setting in Tanzania. A common link, interestingly, is that staff in both areas were not fully utilized and the results could well be different in areas where case loads are high. At this stage, therefore, we can conclude that in settings where capacity utilization at primary facilities is not fully utilized, IMCI seems to be a good buy.