Health services utilization during terminal illness in Addis Ababa, Ethiopia

Abstract

Objectives We describe modern and alternative health services use in terminal illness of adults, and assess whether utilization patterns of TB/AIDS patients are distinct from those of patients suffering from other illnesses.

Methods Data are from post-mortem interviews with close relatives or caretakers of the deceased. We provide descriptive statistics of health care utilization in adults and discuss their covariates in multivariate analyses.

Results Over 85% of terminally sick patients visited a modern medical facility, but less than 40% spent more than 24 hours in a medical facility and only 25% died in one. Traditional healer (11%) and holy water (46%) visits offer a common treatment and healing alternative, but these visits do not co-vary in any consistent manner with the utilization of modern medical services. In terms of the cause of death, we find a higher contact rate with both modern and alternative medical service providers among TB/AIDS patients compared with those suffering from other medical conditions. The duration of illness seems to account for a good share of that variability. Other covariates of health services utilization are socio-economic status, education and age.

Conclusions The contact rate of adults with modern medical facilities in terminal illness is almost universal, but their usage intensity is rather low. Alternative curative options are less commonly used, and do not exclude modern health services use. This suggests that both types of services are considered complements rather than alternatives for each other. Because the contact rate with health service providers is greatest for TB/AIDS patients, it is unlikely that HIV/AIDS-related stigma is an impediment to seeking care. We cannot exclude, however, that it delays health-seeking behaviour.

KEY MESSAGES.

• The contact rate with modern medical facilities in terminal illness is almost universal, but the usage intensity is low.

• Holy water and traditional healer visits are common healing and treatment alternatives, but they do not exclude modern or conventional medical services use.

• The contact rate with modern medical services is higher among TB/AIDS patients, but that is in large part due to the extended duration of the illness.

Introduction

Over the last 20 years, urban centres in Ethiopia have been characterized by an increasing burden of HIV/AIDS and chronic illness. As in other developing countries, health services are often provided through several, sometimes overlapping, channels that include both modern and traditional forms of medicine and healing (Beals 1976; Feierman 1981; Kroeger 1983; Hunte and Sultana 1992; Develay et al. 1996; Ngalula et al. 2002; Nyamongo 2002; Hatchett et al. 2004; Case et al. 2005; de-Graft Aikins 2005). In Addis Ababa, the treatment with self-administered medicine (herbal or modern) and the pursuit of other alternative curative options are relatively common, despite the widespread availability of relatively cheap modern medical services (Kloos et al. 1987; Gedif and Hahn 2002).

In this study, we describe modern and alternative health services utilization in the terminal illness of adults, and investigate their covariates in multivariate analyses. We define modern health services utilization as visits to a hospital, clinic or health centre, and consider visits to a traditional healer and holy water sources as the alternative curative options. Because it has often been argued that the choice of health service is conditioned by the type of disease and concepts of its etiology (Kroeger 1983), we are interested in possible differences in health services utilization by cause of death. For some conditions such as HIV/AIDS, stigma is often also considered an impediment to accessing health care (Awusabo-Asare and Anarfi 1997; Admassu 2000; Valdiserri 2002; Kalichman and Simbayi 2003). Our study also provides a baseline for identifying change in health services use following the rollout of antiretroviral therapy.

Methods

The study was carried out in Addis Ababa. Like many urban areas in the region, it is severely affected by the HIV/AIDS epidemic. HIV prevalence was estimated at 11.7% for 2005 (MOH 2005), or 5.0% according to the 2005 Demographic and Health Survey (CSA and ORC Macro 2006). A limited number of AIDS patients have been receiving ART through the informal market since 2000 (Kloos et al. 2007) and usually at very high cost. In July 2003, the Ethiopian government launched a programme for the provision of antiretrovirals through a co-pay scheme whereby the cost to the patient was US$28 per month (ARC 2005). Enrolment figures have expanded rapidly since 2005, when ART became available free of charge. By the end of 2007, over 40 000 patients had ever initiated ART in any of Addis Ababa's health facilities, and close to 28 000 were receiving treatment. Over 90% of patients on treatment are adults (MOH-HAPCO 2007). This paper, however, reflects the situation prior to the widespread availability of antiretroviral therapy in Ethiopia.

The data come from post-mortem interviews with relatives and/or primary caretakers of the deceased. In these interviews, also referred to as verbal autopsies (VAs), the caretakers were queried about the signs and symptoms the deceased suffered from during their terminal illness (Soleman et al. 2006). Important for this study, questions were included about their health-seeking behaviour. The VAs were administered for adult deaths (12 years and above) randomly selected from deaths recorded in an ongoing surveillance of burials in Addis Ababa. This surveillance covers all cemeteries of the capital and has been described elsewhere (Sanders et al. 2003; Araya et al. 2004; Reniers et al. 2006). The coverage of the surveillance for adult deaths is estimated at 80% or more (Reniers et al. 2009). Among the sources for under-reporting are the burial of residents beyond the city administration limits, the return of terminally sick migrants to their families for care (Urassa et al. 2001; Clark et al. 2007), the repatriation of bodies for burial, failure of cemetery clerks to register burials, and possibly also illegal burials. The VA sampling frame for this study consists of all deaths that occurred in November and December of 2003, excluding cases for which identifying information (i.e. name, age and address) was incomplete or missing.

Verbal autopsies were conducted by a team of two trained community health workers who visited the household 2 to 4 weeks after the death. Of the 1407 cases selected for a VA interview, 78.6% were completed, 4.5% of the caretakers refused an interview, 13.8% of the households could not be found,¹ and 2.9% of the interviews were not completed for other reasons. Eleven of the completed questionnaires were discarded because interviewers judged the respondent's answer as not truthful.

Causes of death were ascertained by means of physician review. These were done for a random selection of approximately half of the completed VAs. Two physicians independently assigned an underlying cause of death. If the assigned ICD10 code (three digits) for the first two physicians did not match, the VA questionnaire was reviewed by a third physician. If the third assessment did not support either one of the previous diagnoses, the case was settled by consensus. In 11 cases the cause of death remained undetermined. The analyses in this paper are based on the remaining 597 cases with an established cause of death. These were reclassified as external injuries, TB/AIDS deaths, and non-TB/AIDS deaths. External injuries are singled out because they may lead to an atypical contact pattern with health services providers. TB and AIDS deaths were combined because it is often difficult to distinguish the two on the basis of a VA interview.

One of the predictors of health services use is socio-economic status. We defined it in terms of assets ownership, and constructed a cumulative summation scale of the ownership or availability of piped water in the compound; electricity; a telephone or mobile phone; a TV; and a car from the workplace. Ownership of a private car is weighted double that of the other items. Using the 33^rd and 66^th percentiles as cut-offs, households are classified as of ‘below average’, ‘average’ and ‘above average’ socio-economic status.

Our main outcome of interest is health services utilization. This is divided into two groups, namely modern (or conventional) and alternative health services.² We consider three aspects of modern health services utilization: ever visit of a medical facility prior to death (yes/no), the time spent in a medical facility (counted in days), and death in a hospital, clinic or health centre (yes/no). These measure the contact rate as well as intensity of modern health services use. As forms of alternative medicine, we consider the consultation of traditional healers (yes/no) and visits to holy water sources (yes/no). Traditional healers (yebahel medhanit awaqi) are the main providers of alternative, mainly herbal, medical services. The VA questionnaire also contained a question about consulting witch doctors (awaqi), but only five such visits were reported and they are combined with traditional healer visits. Holy water (tsebel) visits are a practice among Orthodox Christians. The Muslim equivalent (zemzem) is not as common and not queried in the post-mortem interviews. The history and range of traditional medical practices in Ethiopia are discussed in greater detail elsewhere (Pankhurst 1990; Kassaye et al. 2006).

We first present frequency tables, and to analyse the covariates of health services utilization, we use logistic regression models for the binary outcomes discussed above. To model the number of days spent in a hospital prior to death, we use a negative binomial regression model.³ As a word of caution, the reader should be aware that we describe health services utilization in terminally ill patients. Our sample is thus conditioned on unsuccessful treatment. The health-seeking behaviour of patients who survived spells of illness is possibly different.

Results

The proportion of adult deaths attributed to TB/AIDS is 44%. In the age group 20–64 years, the proportion of TB/AIDS deaths is 55%, which is at the lower end of a range of estimates for 2001 (Reniers et al. 2006). AIDS deaths occur on average at a younger age (38.3 years) than deaths from other medical causes (58.8 years) (Table 1). The relationship between educational status and TB/AIDS mortality is not immediately obvious, but in terms of the socio-economic status of the household, poor households are disproportionately affected by AIDS mortality.⁴

Table 1.

Socio-demographic background characteristics of the deceased by cause of death, Addis Ababa (2003)

	Cause of death
	TB/AIDS (N = 262)	Injuries (N = 40)	Other (N = 295)	(Missing)N
Age				(0)
Mean age	38.3	37.3	58.8	597
Sex				(0)
Women	44.5	3.8	51.7	319
Men	43.2	10.1	46.8	278
Education				(5)
No education/literacy programme/religious school	28.3	4.0	67.7	223
Some/complete primary	52.9	9.6	37.5	136
Some secondary or higher	53.7	7.7	38.6	233
SES				(1)
Below average	56.9	6.9	36.1	144
Average	45.1	6.2	48.8	244
Above average	33.2	7.2	59.6	208

The reported duration of terminal illness is above 6 months for TB/AIDS deaths in about 64% of the cases (Table 2). For other medical causes of death, this is around 51%. Contact with modern health services in terminally ill patients is relatively high: 87% are reported to have ever visited a medical facility (hospital, clinic or health centre) for the condition that eventually led to death. For AIDS deaths this figure is above 96%. Even though these appear to be relatively high values, only 38% of the deceased are reported to have been admitted to a hospital or clinic for more than 1 day. Fewer than 25% spent more than 1 week in a medical facility. The fraction that is reported to have died in a medical facility is under 25% as well, a figure that is close to an estimate for the year 2001 (Reniers et al. 2005). The majority (86%) of health facility deaths occurred at public facilities (not shown).

Table 2.

Health services use in terminal illness by cause of death, Addis Ababa (2003), in percentages

Cause of death	Ill ≤ 1 month	Ill > 6 months	Ever visit medical facility	Spent >24hrs in a medical facility	Medical facility deaths	Ever visit traditional healer	Ever visit holy water source	Exclusively visited traditional healer/holy water
TB/AIDS	6.6	64.2	96.6	40.5	19.5	18.3	51.2	2.3
Injuries	78.8	18.2	65.7	17.7	47.5	3.3	10.0	0.0
Other	29.7	50.7	81.0	39.1	25.8	5.5	44.7	9.7
All	22.2	54.9	87.0	38.4	24.5	11.2	45.8	5.7
N (missing)	576 (21)		585 (12)	562(35)	597 (0)	583 (14)	585 (12)	540(0)a

^aSample limited to cases for whom visits to a modern medical facility, a traditional healer or a holy water source are reported.

Visits to holy water sources as recourse for treatment or healing are reported in 46% of the cases; 11% of the deceased visited a traditional healer. This figure is lower than other reports of alternative medicine use in Ethiopia, but unlike these other studies, our data do not account for self-medication in minor conditions (Kloos et al. 1987; Gedif and Hahn 2002). Both holy water and traditional healer visits are reported more often for TB/AIDS deaths. The proportion visiting a holy water source and a traditional healer is slightly higher for patients who also visited a modern medical facility (not shown). Only 6% of terminally ill patients visited a traditional healer or holy water source without any reported contact with a modern medical facility. The fraction is even lower for TB/AIDS patients.

To explore the covariates of medical services utilization further, we turn to regression models (Tables 3 and 4). For each outcome, we first report bivariate or unadjusted odds ratios (OR) or incidence rate ratios (IRR), and in the second model the net effects after controlling for the other variables listed in the table.

Table 3.

Regression models of the utilization of modern medical facilities (MF) in terminal illness, Addis Ababa (2003)

	Ever visit a MF		Days admitted to MF		Death at MF
	Unadjusted OR	Adjusted OR	Unadjusted IRR	Adjusted IRR	Unadjusted OR	Adjusted OR
	(z score)	(z score)	(z score)	(z score)	(z score)	(z score)
Sex
Female	ref.	ref.	ref.	ref.	ref.	ref.
Male	0.890 (0.47)	0.708 (1.02)	0.878 (0.57)	1.057 (0.22)	1.244 (1.15)	0.944 (0.26)
Age
Age	1.105*** (3.24)	1.065* (1.72)	1.089*** (3.01)	1.052* (1.65)	1.046 (1.62)	1.032 (1.01)
Age squared	0.999*** (4.42)	0.999*** (2.98)	0.999*** (3.67)	0.999*** (2.64)	0.999** (2.28)	0.999* (1.87)
Duration of illness
< 1 month	ref.	ref.	ref.	ref.	ref.	ref.
1–6 months	10.545*** (5.12)	5.994*** (3.30)	4.802*** (4.69)	3.949*** (3.67)	1.077 (0.27)	1.381 (0.94)
> 6 months	5.831*** (6.23)	4.745*** (4.17)	6.583*** (6.66)	5.373*** (5.02)	0.758 (1.14)	1.247 (0.72)
Cause of death
Other	ref.	ref.	ref.	ref.	ref.	ref.
Injuries	0.450** (2.06)	0.427 (1.40)	0.230*** (2.99)	0.566 (0.90)	2.607*** (2.79)	1.763 (1.17)
TB/AIDS	6.581*** (5.08)	2.091 (1.59)	0.975 (0.11)	0.916 (0.33)	0.696* (1.76)	0.434*** (3.28)
Socio-economic status
Below average	ref.	ref.	ref.	ref.	ref.	ref.
Average	1.029 (0.09)	0.981 (0.05)	1.254 (0.79)	1.114 (0.34)	0.770 (1.00)	0.609* (1.70)
Above average	1.140 (0.40)	2.454* (1.94)	2.607*** (3.26)	2.486*** (2.80)	1.775** (2.31)	1.591 (1.60)
Education
None	ref.	ref.	ref.	ref.	ref.	ref.
Primary	2.749*** (2.93)	0.967 (0.07)	0.647 (1.44)	0.442** (2.22)	2.485*** (3.43)	1.589 (1.46)
Secondary	3.444*** (4.11)	1.423 (0.72)	0.952 (0.19)	0.863 (0.44)	2.523*** (3.92)	1.474 (1.26)
Alternative medicine
Holy water visit	1.286 (0.98)	0.815 (0.61)	0.821 (0.86)	1.011 (0.05)	0.641** (2.24)	0.690 (1.64)
Traditional healer visit	1.440 (0.81)	0.401* (1.77)	1.250 (0.62)	1.044 (0.12)	0.690 (1.11)	0.714 (0.94)
Observations		549		528		558
LL(df)		−144.23(13)		−1312.42(13)		−273.81(13)

*P ≤ 0.10; **P ≤ 0.05; ***P ≤ 0.01; z-score in parentheses.

Table 4.

Regression models of the utilization of alternative health services in terminal illness, Addis Ababa (2003)

	Traditional healer		Holy water
	Unadjusted OR	Adjusted OR	Unadjusted OR	Adjusted OR
	(z score)	(z score)	(z score)	(z score)
Sex
Female	ref.	ref.	ref.	ref.
Male	1.159 (0.56)	1.119 (0.38)	0.561*** (3.43)	0.673** (2.03)
Age
Age	1.122** (2.37)	1.127** (2.35)	0.973 (1.38)	0.929*** (2.94)
Age squared	0.999*** (2.62)	0.999** (2.46)	1.000 (1.13)	1.000** (2.24)
Duration of illness
< 1 month	ref.	ref.	ref.	ref.
1–6 months	3.718** (2.27)	2.297 (1.34)	1.617* (1.79)	1.551 (1.38)
> 6 months	4.774*** (2.93)	3.027* (1.92)	3.169*** (5.01)	2.975*** (4.00)
Cause of death
Other	ref.	ref.	ref.	ref.
Injuries	0.593(0.50)	0.999 (0.00)	0.137*** (3.20)	0.157*** (2.75)
TB/AIDS	3.855*** (4.46)	2.990*** (3.04)	1.295 (1.51)	0.882 (0.55)
Socio-economic status
Below average	ref.	ref.	ref.	ref.
Average	0.489** (2.13)	0.603 (1.40)	1.407 (1.60)	1.849** (2.49)
Above average	0.816 (0.64)	1.305 (0.71)	0.981 (0.09)	1.407 (1.26)
Education
None	ref.	ref.	ref.	ref.
Primary	1.946** (1.99)	1.067 (0.17)	1.064 (0.28)	0.844 (0.59)
Secondary	1.200 (0.57)	0.698 (0.88)	0.831 (0.97)	0.506** (2.45)
Religiona
Orthodox	1.702 (0.99)	1.611 (0.83)	5.657*** (4.42)	7.623*** (4.88)
Other	ref.	ref.	ref.	ref.
Utilization of other medical services
Holy water/traditional healer	1.526 (1.60)	1.314 (0.91)	1.526 (1.60)	1.321 (0.94)
Modern medical facility	1.440 (0.81)	0.447 (1.54)	1.286 (0.98)	0.931 (0.21)
Observations		549		549
LL(df)		−175.51(14)		−334.83(14)

*P ≤ 0.10; **P ≤ 0.05; ***P ≤ 0.01; z-score in parentheses.

^aJust over 90% of the deceased in the sample were Orthodox Christians. The remaining 10% includes Muslims, Protestants and Catholics.

Among the significant bivariate predictors of having ever visited a modern medical facility are age, the duration of illness, the cause of death and educational attainment. In terms of the illness duration, however, we cannot exclude a reverse-causal effect because the utilization of health services may prolong life, and hence the duration of the illness episode. The effect of age follows an inverse u-shaped pattern with the odds reaching a maximum at age 36. Declining health care utilization in old age has previously been observed for minor conditions as well as in terminal illness (Kloos et al. 1987; Fantahun and Degu 2003; Case et al. 2005). The odds of having visited a medical facility are highest among those who have been ill for more than 1 month. Possibly because the duration of terminal illness in TB/AIDS cases tends to be relatively long, the odds of visiting a medical facility are greater for TB/AIDS patients than for other causes of death. The odds that a sick person visits a modern medical facility are also greater for patients with higher educational attainment.

After controlling for potential confounders, TB/AIDS patients are still more than twice as likely as other patients to have ever visited a medical facility, but there is an important reduction from the bivariate effect and it is no longer statistically significant. Statistical control for the duration of illness accounts for most of the reduction in the effect of the cause of death (not shown). This suggests that TB/AIDS patients are more likely to visit a medical facility primarily, but not exclusively, because of the longer duration of their illness. The duration of illness itself is the most important predictor of modern medical services use. Education lost statistical significance in the model with controls while the effect of socio-economic status seems to have gained strength. While there was no bivariate association between holy water or traditional healer visits and having been to a modern medical facility, the parameter for traditional healer visits in the model with controls is negative and marginally statistically significant.

The most important predictors for the time a patient spends in a health facility are similar to those for having ever visited a medical facility. The patients’ age, the duration of illness and the socio-economic status of the household correlate with the duration of admission in a similar way as they did for the previous outcome. While odds of visiting a medical facility were higher for TB/AIDS related deaths, no relationship exists with the duration of admission.

The relationship between the cause of death and the indicator of medical services use reverses when we consider the place of death as an outcome: while TB/AIDS patients are more than twice as likely than others to have visited a medical facility, they are only half as likely to die in one. A similar phenomenon has been observed in Mwanza District in Tanzania (Ngalula et al. 2002). As is the case in the regression models of the other outcomes, there is a curvilinear effect of age: the odds of dying in a hospital reach a maximum at age 29 and decline in older age.

The effect of socio-economic status is persistent for the three outcomes: sick people from richer households seek medical care more often and with a greater intensity than those from poor families. Higher levels of education also increase the likelihood of a medical facility death, but that effect loses statistical significance in the model with controls. Sex is not a significant covariate of any of the outcomes investigated in Table 3. In reading these results, one should be aware that the statistical power for many of these tests is fairly limited because of the small sample size. In the full sample of VAs, for example, the bivariate odds that a man dies in a hospital are 44% higher than for women (P = 0.015).

The utilization of modern health services does not co-vary in a consistent manner with visits to a traditional healer or a holy water source, confirming that they should not be considered alternatives to modern medical care. If that were the case, we would find strong and significant negative effects. However, we do find that death in a medical facility is significantly less likely for those who visited a holy water source.

In Table 4, we further investigate determinants of traditional healer visits. The effects of age, duration of illness and cause of death are similar to those for the outcomes related to the utilization of modern medicine. Even after controlling for potential confounders, TB/AIDS patients are more likely to have visited a traditional healer. A similar pattern was identified in Tanzania (Ngalula et al. 2002).

With the exception of the duration of illness, the covariates of holy water visits tend to deviate more from the other forms of medical services utilization. The most important predictors in this case are female gender and Orthodox Christian religion. The latter should not come as a surprise as holy water sources are usually associated with Orthodox Christian churches. Holy water visits are also less common among the better educated portion of the population. The age pattern of holy water visits also contrasts with that of the utilization of medical services discussed so far. In this case, the frequency of visits increases in older age. Holy water and traditional healer visits appear to be positively correlated, but none of the associations reach statistical significance.

Discussion

Over 85% of all terminally ill patients in Addis Ababa visit a modern medical facility prior to death. For AIDS patients the figure is even above 95%. These values are of the same order of magnitude as those observed in KwaZulu-Natal (Case et al. 2005). Having ever visited a medical facility, however, is a crude indicator of services utilization as it reveals little about the frequency or intensity of visits, the stage at which medical care is sought, or the order in which different treatment options are tried out. Some of this dynamic is illustrated by other indicators that we considered. Less than 40% of terminally ill patients spent more than 24 hours at a hospital or clinic. Deaths in medical facilities are even less common (24.5%). Despite a high contact rate, the intensity of health services utilization is thus rather low and this suggests that providers only have a small window of opportunity for case management and treatment.

Holy water (46%) and traditional healer visits (11%) are common treatment and healing options as well, but their level of utilization does not compare with that of modern medicine. Holy water visits are a recourse offered by the Ethiopian Orthodox Church and thus very specific for the setting studied here. Contact with traditional healers is in-between the levels observed in Mwanza District, Tanzania and KwaZulu-Natal, South Africa (Ngalula et al. 2002; Case et al. 2005). More important is that the utilization of these alternative curative options does not seem to co-vary in any consistent manner with the utilization of modern medical services, suggesting that they are used as complements for each other rather than alternatives. This confirms findings from a study in KwaZulu-Natal (Case et al. 2005). The only exception is that those who visited a holy water source are less likely to have died in a medical facility. It is plausible that patients with bleak prognosis are discharged from medical facilities and subsequently revert to more miraculous forms of healing. This hypothesis should be revisited with more detailed information on the timing and sequence of health services use.

In terms of the variation in health services utilization by cause of death, we find that the contact rate with modern medical services as well as traditional healers is higher for those who suffer from TB/AIDS related conditions and those who are sick for longer periods of time. Because of the covariance between cause of death and duration of terminal illness, however, the net effect of the cause of death is much smaller. Nevertheless, our results suggest that the contact rate of HIV/AIDS patients with modern medical facilities is similar to, if not higher than, that for other conditions. HIV/AIDS-related stigma, as it is sometimes suggested, is thus not an impediment to seeking care. However, we cannot exclude that stigma delays health-seeking behaviour. The order in which treatment alternatives are explored cannot be fully appreciated with these data either, but they do suggest that patients expand the range of therapy with a longer duration of illness, and possibly also as their condition deteriorates (see also Feierman 1981; Kroeger 1983; Develay et al. 1996; Ngalula et al. 2002).

Throughout this paper, we have made a distinction between the contact rate and the intensity of health services utilization, and these demonstrate a clear pattern by cause of death. While the contact rate with medical facilities is higher for TB/AIDS patients, their admission time is comparable with that of patients who suffered from other medical conditions. The probability of a medical facility death is even lower for TB/AIDS cases than for patients who suffered from other conditions. The most plausible explanation is that terminal AIDS patients are discharged once physicians (or the patients themselves) realize that the prospects for improvement are grim.

The age pattern of medical services use suggests that teenagers and the elderly were the least likely to visit a modern medical facility. In interpreting the age effect, we should be aware that it may be picking up differences in the cause of death structure in early and late adulthood that are not captured by distinguishing TB/AIDS deaths from other causes. Heart attack and stroke, for example, are more common in late adulthood and may result in an immediate death and thus a small window of opportunity to seek medical care.⁵ Other potential reasons for lower health care utilization in older age are the lack of mobility in the elderly (Kloos et al. 1987) or the unequal distribution of household resources for health care. While consulting a traditional healer follows a similar age pattern as modern medical services, holy water visits do not: the elderly are the most likely to turn to holy water sources for curative purposes. This effect is possibly due to age or generational differences in religiosity.

Education and other measures of socio-economic status co-vary with modern health services use in predictable ways: wealthier and better educated patients have a higher contact rate and greater reported usage intensity. Comparable results have been reported elsewhere (Kroeger 1983; Kloos et al. 1987; Develay et al. 1996; Van der Meer et al. 1996; Fantahun and Degu 2003; Buor 2004; Ahmed et al. 2005). In Ethiopia, most public medical services are cheap; for the poorest even free (Tassew et al. 1996; Haile Mariam and Kloos 2006). Whereas it remains possible that health services costs deter the less well off, the socio-economic gradient in non-use may also arise from indirect or opportunity costs of health services use (e.g. transportation) (Kroeger 1983), or because of non-pecuniary demand-side barriers to accessing care that correlate with socio-economic status (e.g. information, household preferences) (Ensor and Cooper 2004).

An aspect of health services utilization that was beyond the reach of this study, and that we believe should be dealt with more appropriately in future studies, concerns the timing of first seeking medical care, and the sequence in which care and treatment from different providers is sought. Both may be important determinants of treatment outcomes. For several conditions, including malaria and AIDS, the first line of treatment is often self-medication. Failure of these remedies triggers the exploration of other treatment options, which may or may not include traditional healer visits (Nyamongo 2002; Hatchett et al. 2004). The time lost in that process can be crucial. In the case of AIDS, for example, several studies have indicated that late enrolment is one of the main determinants of mortality in antiretroviral therapy programmes (Braitstein et al. 2006; Lawn et al. 2006).