Methods and sources of data

A large part of the information presented in this chapter comes from data originating from the routine information system of the Bwamanda hospital.

The insurance status of all admissions for the 12 month period April 1993-March 1994 was checked through a retrospective study of the existing routine hospital registers of each hospital department¹⁰ (paediatrics, maternity, gynaecology, internal medicine, surgery and intensive care). These registers also record the health centre area of origin for each admission. The existing system of family files, opened after a complete district census carried out in 1985-1986, provides a denominator for each health centre area. This figure was adapted every year so as to account for population growth. In addition, yearly activity reports of the Bwamanda hospital provided detailed information on the number of people in each health centre area that sub- scribed to the hospital insurance scheme. Hence, health centre area specific admission rates, per ward, could be calculated for insured and non-insured patients respectively, allowing for a spatial analysis of utilisation. Statistical significance can be assessed using the Wilcoxon signed ranks test since the differences observed between admission rates in insured and non-insured populations can be meaningfully ranked.

10 Patients from the trypanosomiasis ward were not included in this analysis because most of them come from surrounding districts. Most of the patients admitted in the intensive care ward are in fact transferred to other wards after a few days these admissions are thus counted twice and the real number of admissions is therefore somewhat less.

This spatial analysis was carried out for 17 health centres areas out of the 22 in the district. Five health centres areas were not considered in this analysis: the two health centres located in Bwamanda township with very mobile populations, and three health centres situated at the very edges of the district along a major communication axis In these areas, people living outside the district often claim to live in that health centre area in order to be eligible to subscribe to the scheme. Hence, substantial contamination of the numerators of the admission rates was very likely.

The average length of stay was checked in both insured and non-insured inpatient-population admitted in the 4-month period January to April 1996 (departments of internal medicine, surgery, gynaecology and paediatrics). This data may indicate possible inefficiencies in hospital use by the insured, i.e. a phenomenon of moral hazard.

In addition to this spatial analysis per hospital department, the analysis distinguishes two types of hospital utilisation and justified non-priority hospital utilisation and justified non-priority hospital utilisation, using a set of specific health problems as tracers. The tracer concept was borrowed from the formal sciences (i.e. endocrinology) and has been developed into a method to evaluate the strengths and weaknesses of a health service network (Kessner and Kalk 1973). This methodology leads directly to conclusions about unmet needs for care and services (Carr and Wolfe 1976), and has been used to evaluate the performance of health care programmes in both industrialised (Buekens 1984) and developing countries (Amonoo-Lartson and De Vries 1981, Carr et al 1989; Pangu 1988). Tracers need to fit the following criteria: they should be sufficiently prevalent; their diagnosis should be relatively easy; they should have a significant functional impact which can be influenced by patient care (i.e. the health problem is vulnerable to a technical intervention by the health services system). In this case the tracers should make it possible to test whether the distribution of hospital utilisation corresponds to the distribution of needs within the district.

Patients who underwent surgery for uncomplicated hernias were used as tracers for justified non-priority hospital utilisation. Patients (male and female) who underwent surgery for strangulated abdominal hernias (mainly indirect inguinal hernias), and women who had a caesarean section or a symphysiotomy, were identified as tracers of justified priority hospital utilisation. In addition, the indication for intervention was systematically checked for all women (whatever their origin) who underwent a caesarean section/symphysiotomy in the period 1993-1996. A distinction was made between interventions done for absolute maternal indications and interventions done for non-absolute maternal indications. This distinction gives an indication on the extent, if any, of supply-induced over-intervention (De Brouwere et al. 1996). The following situations were classified under the heading absolute maternal indications: brow presentation, ante-partum haemorrhage, transverse lie and shoulder presentation, foeto-pelvic disproportion, ruptured uterus and postpartum haemorrhage that led to hysterectomy.

The data on the different tracers do not all cover the same time periods. In order to have sufficiently large figures, hospital use for hernias was analysed over a 3-year period (April 1993 - March 1996) and caesarean sections/symphysiotomies over a 5-year period (April 1991- March 1996). This is not considered a problem because of two reasons. First, the fact that except for the years 1992 and 1994, the yearly subscription rates to the scheme were remained relatively constant in the period 1988-1996. Second, because it was established that the cohort of people who joined the scheme in 1988 remained almost identical in the 9 subsequent years (unpublished data).

The health centre area specific admission rates were also used to calculate coefficients of localisation (CL) and location quotients (LQ) which both provide a measure of spatial concentration of the utilisation of the hospital (Joseph and Phillips 1984). A coefficient of localisation measures the concentration across regions of a given phenomenon (in this case, hospital ad- missions), relative to that of a base line. They were calculated for each hospital department, as well as for the three tracer conditions. A CL is calculated as follows:

CL = 1/2

where CL is the coefficient of localisation; Admⁱ the number of admissions from area i, and Pⁱ the population of area i. Values between 0 and 1 reflect increasing levels of localisation. In the present case, the lower the co- efficient of localisation, the more it indicates that admissions coming from a given health centre area are distributed in the same way as the population in the district. A polar value of zero would imply a perfectly equal admission distribution, meaning that the admission-population ratio is the same for all health centre areas (Brown 1994). CL are however statistically not testable (Joseph 1982).

A location quotient facilitates easier and more accurate assessment of inter-area differences in hospital admission It was assessed for the admissions concerning the three tracer conditions. A LQ is calculated as follows:

where LQⁱ is the location quotient for area i; Admⁱ the number of admissions from area i and Pⁱ the population of area i. A value greater than 1,0 means that a health centre area has more admissions than could be expected, that is relative to its share of total population. Conversely, a value less than 1,0 means that an area has been under-serviced. A value of 1,0 means that a health centre area has exactly the number of admissions warranted by its share of total population The comparison of location quotients has to be interpreted in the light of the average admission rate upon which individual LQⁱ values are based. The range of LQⁱ will be presented in the form of box-plots indicating lowest and highest values, median values, and 25^th and 75^th centiles.

The comparative advantages of LQⁱ over CL are twofold. First, LQⁱ measure inter-area differences in admissions in this case, differences between health centre areas whereas CL measures the concentration of admissions across the entire district. Second, LQⁱ allow for statistical testing of medians of the different health centre area admission rates

Other data, not coming from the routine information system, were collected especially for the purpose of this investigation.

In order to further appreciate the importance of non-justified hospital use, if any, by insured and non-insured patients in the different wards, a bed- census was carried out on March 23^rd, 1996. A bed-census, in essence, is a cross-sectional snapshot of the utilisation of hospital beds on one particular day (Buvé and Foster 1995; Pannarunothai 1995). This technique was used in order to assess the proportion of beds that were inappropriately occupied, i.e. bed occupation by patients who could have been treated at lower levels of the health system, and to appreciate whether there was in that respect a difference between insured and non-insured patients. A locally adapted Appropriateness Evaluation Protocol (Gertman and Restuccia 1981) was applied to all 218 patients present in the hospital that very day. The protocol was redesigned with the Bwamanda district health team, based on the format used in a study conducted in Oxford in 1986 (Anderson et al 1988) A list of criteria was established related to medical, nursing and life-support services (Box 1). The patient was considered to have a positive reason to be in the hospital if any criterion was met. A second list consisted of criteria determining why the patient was not at home if none of the positive criteria was applied. The physician in charge of the hospital ward, the head of nursing and an external research person administered this instrument, blind to the insurance status of the admitted patients examined.

Finally, the team filling in the questionnaire during the bed census exercise also addressed the question whether the admission could be considered justified, given the level of functioning of the first line health centres. The objective was to appreciate a difference, if any, between insured and non-insured admissions. This assessment is based on a methodology tested else- where in Congo (Kasongo Project Team 1982), and provides an indication of the efficiency of hospital use. For that purpose, admissions were classified in four categories: (i) the hospital admission was justified; (ii) the patient could have been cared for at health centre level under certain conditions (e.g. introduction of new equipment, drugs, case-management guidelines, etc.); (iii) the patient could have been cared for at health centre level as it functions presently; or (iv) the patient's situation is not clear-cut and doubts remain.

Box 1. The Appropriateness evaluation form