Illuminating cases: understanding the economic burden of illness through case study household research

Abstract

Understanding the economic burden of illness for households can inform pro-poor health and social protection policy, yet research is in its infancy and appropriate methods require further debate. Quantitative studies are powerful when applied to the right health policy questions, including the measurement of illness cost burden indicators. However, this paper argues that not all dimensions of economic burden can be measured easily, some dimensions relevant to policy, such as social actors' responses to illness and their strategies to cope with illness costs, cannot be reduced to quantitative indicators at all, and large-scale surveys may overlook context-specific processes operating at household level that influence people's paths in and out of poverty as a result of illness. This leaves scope for longitudinal case-study household research to enhance understanding of economic burden and provide additional policy insights on how to better protect households from cost burdens and improve resilience. Drawing on the experience of research in urban Sri Lanka, the paper sets out several comparative advantages of case study research in this area. First, it complemented household survey data by revealing the complex and dynamic nature of illness costs and how these cost patterns (for example, sudden cost peaks) influenced household ability to manage costs. Secondly, it improved understanding of vulnerability or resilience to illness costs by capturing the diverse resources, within and outside the household, used by people to cope with illness costs, and the social institutions and decision-making processes that influenced access to them. Thirdly, the cases enabled the research to develop a picture of the inter-connected factors mediating the impact of illness on livelihood outcomes.

Introduction

Clearly one good case can illuminate the working of a social system

            (Gluckman 1961: 9)

Ill health can cause household impoverishment through income losses and medical expenses that trigger a spiral of asset depletion, indebtedness and cuts to essential consumption (Gilson 1988; Pryer 1989; Haines et al. 2000; Kabir et al. 2000), processes brought into sharper focus by the socio-economic impact of HIV/AIDS. With poverty reduction at the centre of the development community's targets and strategies, understanding the economic burden of illness for households, and the policy arrangements that exacerbate or mitigate this burden, can inform pro-poor health policies.

Economic burden is defined here as expenditure on seeking treatment (direct cost), production and income losses (indirect cost), related coping strategies, and their consequences for the household livelihood in terms of indicators such as the number of workers and working days, asset portfolios, income and food consumption levels (Scoones 1998).

HIV/AIDS, malaria and other diseases pose a serious threat to poverty reduction, yet research in this area is in its infancy, and appropriate methods to research the economic burden of illness and factors mitigating impact require further debate and refinement (Booysen and Arntz 2003). The dominant research design in the public health and health policy arenas is quantitative and experimental, the gold standard against which other designs are judged (Baum 1995; Victora et al. 2004).

Without doubt, quantitative studies are powerful when applied to the right health policy questions, including the measurement of certain economic burden indicators such as the direct and indirect costs of illness. However, this paper argues that not all dimensions of economic burden can be measured easily, some dimensions relevant to policy, such as asset strategies mediating the impact of illness on livelihoods, cannot be reduced to quantitative indicators at all, and large-scale population-based analysis can overlook context-specific processes operating at household level that influence people's paths in and out of poverty as a result of illness. This leaves scope for case study household research that can adopt a more naturalistic methodology, seeking to understand the economic burden of illness as it unfolds within the context of a household, and the ways that people in their everyday lives take action to treat illness and cope with its costs.

The paper makes the case for research designs that aim for depth of understanding as well as breadth, arguing that more emphasis should be placed on case study work to complement survey results, but also as an approach in its own right that can generate additional understanding of the factors influencing vulnerability to illness costs, such as income insecurity or social networks that provide support in paying for treatment. This broadened toolkit can therefore generate policy-relevant insights missed by conventional quantitative studies.

Debates on methodological complementarity are hardly new to public health, most notably between anthropologists and epidemiologists whose collaborations have enriched understanding and interventions (Baum 1995; Inhorn 1995; Trostle and Sommerfield 1996). But here the case for a specific research design is made for understanding the economic burden of illness for households, one which takes a household survey method traditionally associated with disciplines such as economics and public health, and adds a complementary case study approach derived from the discipline of anthropology. Case studies are not designed to be statistically representative, but aim to strengthen understanding of complex realities and household processes that cannot be measured easily, so the concern is not sample size but the depth of understanding generated (Mitchell 1983; Yin 1994; Seeley et al. 1995; Thomas 1998; Coast 1999).

To demonstrate the benefits of case study research, the paper draws on work undertaken in urban Sri Lanka which combined a cross-sectional household survey with longitudinal case study research. The next section examines different concepts and indicators that can be used to research economic burden, broadly grouped as cost burdens, asset strategies and livelihood outcomes, and the three subsequent sections discuss the advantages case studies bring to researching these concept and indicator categories.

Researching economic burden: concepts and indicators

Income and expenditure-based concepts and indicators

Conventional economics conceptualizes and measures poverty in terms of income and expenditure (consumption). Using this approach the economic burden of illness can be measured using two related indicators.

(1) Health care expenditure as a proportion of household income (direct cost burden)

This indicator reflects concern about the opportunity cost of health care spending, or its potential consequences for household members' ability to meet other basic needs (Russell 1996). Recent studies have classified health care payments above 10% of income as ‘catastrophic’ for households, assuming that above this threshold payments are likely to cause cuts to food consumption, debt and impoverishment (Prescott 1999; Ranson 2002). A more refined indicator changes the income denominator to that remaining after basic consumption needs have been met (capacity to pay) (WHO 2000). A health expenditure burden greater than 40 or 50% of capacity to pay is assumed to be ‘catastrophic’ for households (WHO 2000; Wagstaff and van Doorslaer 2003; Xu et al. 2003).

(2) Production and income loss caused by illness as a proportion of ‘normal’ income (indirect cost burden)

Income losses caused by illness, particularly serious and prolonged illness, are often a more significant cause of impoverishment than direct costs, undermining household members' command over essential goods and services.

Asset and vulnerability-based concepts and indicators

Income-based poverty approaches have been criticized for not capturing the range of resources and strategies that people mobilize to access commodities and cope with shocks such as illness. Sen's important work on entitlements, for example, established that an individual's access to commodities, or their ‘entitlement set’, is determined not just by income but by a range of production, exchange and transfer processes including government services (Sen 1981; Drèze and Sen 1989). The asset portfolios at people's disposal, including policy-derived resources and less tangible assets like social relationships, also influence their ability to cope, or their vulnerability or resilience to shocks such as illness (Swift 1989; Moser 1998). The concept of vulnerability also makes poverty analysis more dynamic since asset depletion increases vulnerability to future shocks. Applying this more complex and dynamic approach to economic burden requires additional indicators.

(3) Asset portfolios and coping processes

In response to illness costs, household members may adjust consumption or mobilize assets. These responses need to be included in economic burden analysis because they can mitigate the overall burden (for example, a small loan without interest from a friend) or exacerbate it (selling productive assets).

(4) Livelihood outcomes

Illness costs and asset strategies have consequences for the household economy, assessed using indicators such as the number of workers, days worked, material and social assets, income and consumption levels, and vulnerability to future shocks. Analysis can be extended to people's own criteria for livelihood ‘well-being’, such as self-esteem or physical security (Chambers 1995; Scoones 1998).

Using these indicators, the economic burden of illness becomes a complex empirical question of whether households can manage cost burdens over time, in terms of continuing to work, sustaining consumption and preserving assets and self-esteem, or whether they are pushed towards risky strategies that damage asset portfolios, reduce consumption and threaten the sustainability of the household economy and its existence as a social unit (Sauerborn et al. 1996; McIntyre and Thiede 2003; Russell 2004). A simple indicator of economic burden or ‘catastrophic’ illness is therefore difficult to specify. Cost burden indicators (1) and (2) are more easily measured and lend themselves to quantitative research but only capture the potential or likely consequences of illness. Coping and consequence indictors (3) and (4) are harder to measure but capture actual processes leading to impacts.

Most studies in this field have had as their main objective the measurement of the costs of illness (indicator 1 and to a lesser extent indicator 2) and have used conventional economic (quantitative) methods for this purpose. Fewer studies have thoroughly researched the coping and consequence indicators (3 and 4) relevant to economic burden (Russell 1996, 2004). The case study approach can address these gaps in knowledge around indicators 3 and 4, by observing and recording social actors' responses to illness and their coping strategies, and interpreting the complex factors mediating the impact of illness costs on the household livelihood over time. Case studies can also measure cost burden indicators (1 and 2) in more detail and over time, revealing characteristics that might be missed by a survey tool, for example burden fluctuations and peaks that can have serious implications for household budgets, ability to pay and asset strategies.

Policy-relevant research on the economic burden of illness, based on a broader concept of poverty that encompasses vulnerability and asking questions about the four indicators above, requires an interdisciplinary approach to method. A quantitative economic research design conventionally used to measure illness costs can be strengthened through the adoption of a complementary case study approach drawn from anthropology. This approach adopts methods traditionally used by anthropology such as participant observation, in-depth interviews and ethnographic description, but can also be adapted for the specific purposes of economic burden research, recording illness cost indicators and expenditure patterns in greater detail over time.

The Sri Lankan research used these indicators to collect household data between 1998–9 in two low-income areas of the capital, Colombo. The research design, described in more detail elsewhere (Russell, in press), had three components: (1) individual and group interviews to generate qualitative data and inform survey design; (2) a cross-sectional household survey of 423 households (2197 individuals) that produced a profile of household income (expenditure) and assets, acute illness episodes (in the previous 2 weeks) and chronic illnesses, treatment actions, illness costs and coping strategies; and (3) 16 case study households that were purposefully selected for longitudinal study over 8 months using the survey data as a sampling frame. The cases were selected from the survey profile to represent four household income (per capita) quartile groups in the two communities, different degrees of asset vulnerability and different illness experiences. In other words, the case studies were chosen to be ‘typical’ of different household types in the two communities. Each case study household was visited every 2 weeks to record information relevant to the study objectives, using structured and semi-structured interview formats, conversation and observation, and also illness and expenditure diaries completed by household members over short periods.

Measuring direct and indirect cost burdens

Household survey methods are suited to measuring illness cost indicators and their statistical power gives them a comparative advantage over smaller case study samples. Numerous cross-sectional survey studies have measured patient or household direct costs of illness, and to a lesser extent indirect costs, for specific diseases such as HIV, TB and malaria or for all illnesses (for a review see Russell 2004). A few of these survey instruments included questions on income to calculate cost burdens, and one study analyzed large income and expenditure survey data sets from 59 countries to calculate the extent of ‘catastrophic’ health care payments in different health care settings, using a threshold of 40% of capacity to pay (Xu et al. 2003).

The cross-sectional survey used in the Sri Lankan study generated profiles of illness, treatment and cost burdens that were representative of the study population. A simple but important finding, for example, was that the majority of households incurred a low burden of direct illness costs because of access to free public health care, used especially for regular treatment of chronic illnesses and inpatient care: 77% of households experiencing illness (n = 323) incurred a direct cost burden of less than 5% per month and 90% experienced a burden of less than 10%. However, the findings also revealed a gap in the protection offered by public health services, with a large minority (10%) experiencing a direct cost burden above 10% of monthly income. Households from the poorest income quartile experienced higher cost burdens than better-off groups (Russell 2001).

A weakness of cross-sectional surveys is their inability to capture variations in illness costs over time, due to either seasonality or disease progression (Booysen and Arntz 2003). Longitudinal surveys can address this problem (Sauerborn et al. 1996), but not more fundamental measurement or data capture difficulties revealed by the complementary case study data. Some of these difficulties are examined below to show why survey design and analysis require caution and can be informed by and criticized by case study research that can use both quantitative and qualitative methods.

Capturing complex and dynamic cost burden data

In a 1 or 2 hour household survey interview, people are usually unwilling to disclose or unable to recall income data needed for cost burden calculations. Household expenditure is a good income proxy in many contexts, however expenditure data from a survey interview is still subject to considerable strategic and recall bias. Working with a family over an extended period and developing relationships of trust, in contrast, allowed a fuller picture of household members' work, income and expenditure patterns to be built up. Of particular importance for understanding the economic burden of illness were income fluctuations over time (days, weeks and months), which had a considerable bearing on the cost burdens experienced by households. In the poor settlements of Colombo, income variations, caused by illness and the vagaries of casual labour markets, were a common source of livelihood insecurity.

The case studies also revealed characteristics of cost burden numerators (health expenditure and work days lost) that a survey instrument would find difficult to record. First, they could capture cost data from complex illness and treatment events for which a person, or sometimes several people, had used several providers with multiple cost items over an extended period (Box 1).

Secondly, case studies revealed ‘hidden’ costs that might remain undisclosed in a brief survey interview, for example people spent money on religious and ritual-based therapy but were cautious about reporting these costs until familiarity with researchers had been developed. An exploratory approach also meant unusual cost items were captured, notably non-medical direct costs for special foods and transport. For example, following an admission for a hysterectomy operation one family spent more on feeding and transporting visiting relatives over several days than on medical costs. Only a well-designed household survey, combined with an experienced enumerator and sensitive questioning techniques, could capture these less visible items in a short survey visit.

Thirdly, longitudinal case studies captured dramatic fluctuations in health care expenditure and income loss over time that were smoothed by the averages calculated from survey data (Figure 1). Even the monthly peaks in Figure 1 are an average cost burden over 30 days that smooth higher daily cost burden peaks, which expressed as a proportion of a day's wage could easily be over 100%. This peaky or ‘lumpy’ characteristic of illness costs was an important characteristic of its magnitude and economic implications because the poorest and most vulnerable households, dependent on a low daily wage, found it difficult to manage even the costs of common illnesses and usually had to borrow or pawn jewellery to cope (Box 2).

Together, fluctuating income and illness costs meant cost burdens varied dramatically over days, weeks and months, and over an 8-month period the peaks had a significant impact on household budgets and responses. The experience of Pushpa's household (Figure 2 and Box 3) reveals these patterns, which were typical for moderately more resilient households in income quartile 3 (lower vulnerability was linked to job security, jewellery assets and stronger social networks). In Figure 2, month 8 is particularly revealing because illness-related income losses lowered income to just above the local poverty line and direct illness costs absorbed a considerable proportion of remaining income, leaving too little to meet food and fuel needs (see Box 3).

Capturing combined cost burdens

Health care expenditure is one of many demands on a household's budget and cannot be viewed in isolation if a clear picture of the economic burden of illness is to be represented. Lost earnings caused by illness make budget constraints even more serious. In the Sri Lankan study sites, 25% of households (the lowest income quartile) fell below a minimum-needs poverty line of US$21.00 per capita per month, a harsh poverty line (lower than US$1.00 per day) based on the income a household needed to purchase enough food, fuel and soap for its members. Furthermore, about 50% of households (the two lowest quartiles) earned less than US$30.00 per month or a dollar per capita per day. In addition to their struggles for enough food and fuel, these households had to find money for numerous other expenses like health care, education, rites of passage and debt repayments, as well as to finance addictions to tobacco, alcohol or other narcotics.

Case study research was well suited to understanding the implications of illness costs for household budgets and consumption patterns because it allowed researchers to collect detailed expenditure data over time using structured interviews and expenditure diaries filled by adult household members for limited periods. The data showed that among the poorest households a high proportion of income was spent on food (70–90%) and any other expense required borrowing and asset strategies (see Selvaraja in Table 1). Better-off households spent a lower proportion of income on food (50–60%) and had money available to pay for treatment of common acute or chronic illnesses. However, in months when workers lost earnings (see Figure 2 for example) or health care spending coincided with other essential expenses, the combined cost burden was often beyond these households' monthly budget and necessitated asset strategies, illustrated by the case of Pushpa in Table 1. Treatment spending absorbed 10% of household income, but other expenses (education, clothes, debt repayment and notably an obligation to pay a rotating savings group, known as seetu) pushed spending beyond the month's earnings (earnings also reduced by illness – see month 2 in Figure 2). Pushpa had to buy food on credit from the local shop, pawned jewellery and borrowed from a friend (no interest charged).

Capturing sensitive expenditure data

The survey instrument could not collect data on decision-making processes for spending, or on sensitive expenditure items like narcotics. Extended stays in the two communities indicated that narcotic consumption by males was widespread (tobacco, alcohol, heroin, marijuana), and the case studies revealed narcotic spending to be a heavy cost burden for some households, which affected family members' capacity to pay for health care and other essential needs. Among the poorest households for example, even five cigarettes per day cost US$0.40 (Rs.30), a considerable proportion (15–20%) of an average daily wage (US$2.30–3.00). These figures were complemented by qualitative data that revealed the implications of narcotics for impoverishment:

“Nandawathi's husband is drinking and he spends her small earnings from the rice packets she sells… But that is not all. Her son takes heroin and she has to give him Rs.50 a day. They can only eat, nothing else. If there is a marriage or other social event she does not go …”

            (Dilani, woman from case study household in upper income quartile)

Researching household assets and strategies

Household surveys can provide a useful profile of the strategies used by household members to cope with illness costs across communities and can also analyze whether different strategies, such as borrowing from a moneylender at high interest, might be of higher or lower risk to household livelihoods. However, survey instruments cannot capture the full range of household asset portfolios, the decision-making processes influencing their use, or the social and institutional contexts driving these processes. Longitudinal case study research can enhance understanding of these processes and explain patterns observed by a survey, and, if carefully selected, can be the main component of a research design that aims to improve knowledge and develop theory about assets, coping processes and factors influencing vulnerability (see, for example, Seeley et al. 1995; Sauerborn et al. 1996; Baylies 2002).

Capturing complex and less visible assets

The study of a family's interactions with other households and organizations builds up a rich and detailed picture of people's asset portfolios, including less tangible or measurable assets such as organizational skills, access to resources in the community (mediated by institutions) and the size and strength of social networks at times of illness.

Access to financial institutions, notably savings and credit groups, was an important factor influencing people's ability to obtain money to meet the demands of a sudden illness cost. Traditional rotating savings groups (seetu) and NGO-based credit societies were the most common organizations available to people, usually women. However, the case study research revealed that women from the income-poorest households were unwilling to join or often excluded from seetu groups, and to a lesser-extent NGO-based credit societies, because of their inability to make regular payments; other group members could not be sure they could save or repay money. Those with husbands who drank were particularly affected.

Researchers could develop comprehensive data on people's social networks – their size, range and strength, and how, when and why these networks were used for managing illness costs and mitigating impact. Data on social networks were collected through regular interviews and observations, and one special interview devoted to exploring the family's extended family, friends and other actors ‘important to their lives’. A visual tool known as a network map was developed from other anthropological work (Wallman 1984) to help with data collection and to record actors in the network and the support received or given. Qualitative data and field-notes provided important insights into social norms governing association and making claims on networks. Self-esteem and fear of gossip were key factors limiting the claims that could be made on parts of the network, and borrowing from moneylenders and employers outside the community was often a preferred option:

“Have you not heard the gossip of the women in this area? They will say all sorts of things and spread rumours.”

            (Selvaraja, woman from case study household in lowest income quartile)

“(I wouldn't want) … other people talking about us when we walk down the street, saying ‘they haven't paid them back yet’.”

            (Valli, woman from case study household in lowest income quartile)

“My brothers have a lot more money than I have but I don't go to ask because my brothers' wives will talk. Because we take on interest (from a moneylender), there is no problem.”

            (Mayori, woman from case study household in second highest income quartile)

An illustration of the rich data about social networks and the factors influencing their strength and use is shown in Box 4. Data analysis, however, needs to go beyond individual cases to distil common and contrasting concepts and patterns across cases. Qualitative data analysis techniques were used to identify these themes from all interviews, and in particular to understand the interacting factors that explained why network strength varied, for example income and education levels, reputation for returning a favour or repaying a loan, as well as fear of gossip.

Capturing decision-making processes

Knowledge about decision-making processes within the household around issues such as where to go for treatment or coping strategies can be very useful for shaping effective policy responses that are sensitive to access barriers and people's preferences, and so improve service uptake. These (intra-)household processes are better understood through analysis of family relationships, power relations and people's narratives on these subjects.

Case studies are also well suited to researching household processes that mediate the economic burden of illness: decision-making around expenditure, for example, has already been noted. The cases also captured decision-making on asset mobilization and revealed common coping sequences among the studied households that could be distilled into a plausible sequence applicable to all households in this particular setting (Table 2). The sequence reflected people's perceptions of the different costs or risks that strategies posed to their livelihood or well-being; for example, credit from a local shop to purchase food was a lower risk strategy than borrowing from a moneylender or cutting food consumption. The coping strategy adopted also usually indicated a level of vulnerability or resilience, as some households were down to their last assets or options and were forced to take riskier strategies.

Researching livelihood outcomes

Illness costs and coping strategies will cause changes to livelihoods – including impoverishment – defined here as livelihood outcomes when measured at a specific point in time. A longitudinal design is necessary to evaluate these outcomes, particularly for the prolonged impact of chronic and terminal illnesses like HIV/AIDS. Limited knowledge about the socio-economic impact of illness on households and poverty is partly due to the lack of such longitudinal impact studies (Russell 1996; Booysen and Arntz 2003).

Quantitative studies on impact can be strengthened considerably, in principle, by a controlled design that compares affected households with ‘unaffected’ or ‘healthy’ households, so that livelihood changes can be attributed to illness rather than population-wide changes caused, for example, by recession and rising unemployment (Guinness and Alban 2000; Bachmann and Booysen 2004). Only six quantitative studies with a control population were identified, all on the economic impact of HIV/AIDS, two of which used a cross-sectional survey (Pitayanon et al. 1997; Kongsin et al. 2001) and four a longitudinal survey (World Bank 1997: Kagera survey; Menon et al. 1998; Ngalula et al. 2002; Bachmann and Booysen 2004).

In practice, the limited number of controlled studies reflects practical and intellectual difficulties with their use for illness impact studies. First, in the social world, as opposed to a laboratory, a control group of ‘healthy’ or ‘unaffected’ households is unlikely to exist or hard to identify (all households experience illness). If the impact of only serious illnesses or one disease is being investigated, it may be easier to identify ‘unaffected’ households for a quasi-experimental design, but with HIV/AIDS studies questions remain as to whether the control group is ‘unaffected’, since infection may not be disclosed and, more importantly, the disease is so pervasive in some settings that all households will be affected (Bachmann and Booysen 2004). Secondly, it is difficult to recruit control households with matching characteristics in terms of size and composition, livelihood activities or employment security, social networks or service access and utilization.

A third conceptual difficulty with quantitative designs more generally is that livelihood trajectories are influenced by inter-related variables that are hard to isolate into dependent and independent variables for statistical analysis. The causal chain linking illness and livelihood change is not a simple one, but involves a complex chain of numerous inter-related variables that mediate impact. In contrast, case studies are well suited to capturing these complex processes and can complement quantitative designs to reveal the mechanisms that explain the impact patterns observed in ‘affected’ and ‘unaffected’ cohorts.

Capturing multiple events and the dynamic nature of poverty and vulnerability

The Sri Lankan cases revealed the perhaps obvious but often forgotten reality that events other than illness combine to contribute to financial difficulties, such as a marriage, a funeral, a fine, school books, a large debt repayment or a worker's failure to send a remittance. These events needed to be viewed together because each had repercussions for the other, in terms of cash availability and coping capacity over time.

Some events also triggered tension, jealousy, conflict or cooperation in relationships, which in turn influenced levels of support and so exacerbated or mitigated the economic burden and its impact for some individuals. The example of Valli and Siva, on a trajectory of impoverishment for many inter-connected reasons (Box 4), is illustrative of the interacting events over time that contribute to vulnerability and livelihood decline.

The longitudinal case studies also revealed the impact of illness and other shocks on future vulnerability, and so the dynamic nature of poverty. For example, among the poorest households low but frequent illness costs, combined with other expenses, constituted a persistent attack on budgets, assets and livelihood security, already illustrated by the case of Selvaraja (Box 2). The criteria for selecting the case studies and their selection from the survey meant these processes could be considered typical and ‘nested’ within a large group of households from the same community with similar characteristics of income-poverty, budget constraint and asset vulnerability: in the Sri Lankan case, at least 25% of households struggled under similar budget constraints.

Capturing continuity of processes linking events and outcomes over time

One of the methods researchers can use with case studies is life history analysis that can document, retrospectively, long-term livelihood trajectories and their influence on the more recent livelihood paths being recorded by prospective research. Previous circumstances and events (including serious illness) place households on trajectories characterized by struggling, coping or improving, and in many cases the illness cost burdens recorded during research are unlikely to make a major difference to these paths unless they are high or persistent, although they can exacerbate existing vulnerability and struggling. For example, three of the Sri Lankan case study households were on trajectories of livelihood decline and impoverishment before research started, due to earlier serious illness that had stopped the main breadwinner from working and caused considerable asset depletion and indebtedness.

When researchers live with or close to households, which is possible with case study work, there is the added advantage of capturing the continuity of processes linking events and outcomes over time. This might be termed a ‘diachronic’ perspective that views why and how an illness event has an impact on the outcome indicators recorded. In contrast, a longitudinal survey that measures household states in a series of separate points over time is very useful, but provides only a synchronic view with limited ethnographic data on the decisions and mechanisms that explain changing household outcomes.

Listening to research subjects' own perspectives on livelihood change

The livelihood impact of an illness extends beyond work, assets and consumption, since a person's standard of living and quality of life is multi-dimensional. Anthropological and ‘participatory’ approaches to poverty analysis have demonstrated that people's own (emic) concepts of well-being or disadvantage differ from those of outsiders or ‘experts’ (etic) (Chambers 1995). Similarly, academic thinking about poverty has shifted from income-based approaches to concepts of vulnerability and capability, the former giving importance to questions of power, voice and security, the latter to people's functionings, which include not only people's ‘doings’ but their ‘beings’ – their identity, status and self-esteem (Drèze and Sen 1989; Chambers 1995; Scoones 1998; World Bank 2000).

While surveys can measure more conventional material, livelihood outcomes, case study and qualitative designs are better suited to assessing these more subjective outcome indicators. The work in Sri Lanka could record the anxiety or distress that accompanied sickness and resulting financial difficulties, or the low self-esteem felt by those who were in debt to friends and family.

People's narratives were also used to triangulate findings about livelihood change from other sources, for example confirming the income and food consumption declines already recorded through structured questions:

“If I am ill there is no money for food … and my legs are hurting and I couldn't go to work today … we didn't eat this morning but I will try and get some food today and cook this evening … somehow… I have not been cooking – we only eat a proper meal once per day … even during Pongol (Hindu festival) we had no food to eat.”

            (Valli, woman from case study household in lowest income quartile)

Conclusion

This paper does not question the strengths of longitudinal quantitative and controlled designs for research on the economic burden of illness and poverty. They give statistical power and strengthen the claims that can be made about the impact of illness on measurable household characteristics like cost burden, (un)employment, income and consumption levels, informing important policy questions about service coverage, equity and the impact of health policy measures. For example, measuring the incidence of ‘catastrophic’ health care payments across populations in poor countries can identify vulnerable groups and gaps in health service protection (Wagstaff 2002; Wagstaff and van Doorslaer 2003; Xu et al. 2003), and these findings can be communicated in simple but forceful ways to policy-makers (White 1984).

Questions have been raised, however, about the feasibility or appropriateness of measuring all aspects of economic burden and the processes mediating the impact of illness on household poverty. Not all can be reduced to numbers. Case study research views illness events and responses within the household more holistically, focusing on the inter-connectedness of events, how these shape responses, and their impact on the household economy. It allows a range of methods to be used to collect complementary data and triangulate findings, and facilitates observation to compare what people do with what they say they do.

With these advantages, the intensive study of a small number of families in Sri Lanka demonstrated how a case study approach can generate additional policy-relevant knowledge that would be missed by more conventional survey approaches. First, the more detailed cost burden and related asset or coping data inform health financing debates around the pros and cons of free health care as opposed to user charges. In the Sri Lankan setting, the case study findings showed that free health care services protected the poor from high direct cost burdens, and so over time protected assets and mitigated or prevented livelihood decline. Case study data also showed that policy-makers relying on survey figures may underestimate illness cost burdens and their implications for impoverishment:

• burdens usually peaked over short periods of a few days or a week, making costs harder to manage;

• certain types of illness cost were ‘hidden’;

• combined cost burdens arising from a range of essential expenses meant poor households had no capacity to pay for health care, and had to borrow, pawn jewellery or cut consumption; and

• ‘small’ but frequently incurred burdens were a persistent attack on household budgets, assets and livelihood sustainability.

These data inform policy concerns about what constitutes a ‘catastrophic’ cost burden, since a burden of 5% or even less could trigger asset depletion for the poorest and most vulnerable case study households, but a higher burden was not so ‘catastrophic’ for those with stronger asset portfolios that enabled them to cope. Case study cost burden data therefore improved interpretation of household survey data that described cost burdens across the whole population and for different socio-economic groups. Furthermore, by understanding people's responses to cost burdens and how these filter livelihood impacts, policy-makers become better informed about the implications of their health service financing and delivery decisions.

Secondly, the case study approach enabled detailed demand-side analysis of treatment behaviour for acute and chronic conditions, and could follow up the implications of treatment decisions for direct cost burdens, providing policy insights for pro-poor health service delivery measures. Public service delivery weaknesses at lower levels of the health system, for example, discouraged uptake of public services for common acute illnesses and pushed people, even from the poorest households, towards the private sector, increasing direct costs and undermining public service protection (Russell, in press). These delivery weaknesses included long waiting-times and poor inter-personal quality, particularly a lack of patient focus in doctor-patient interactions (Russell, in press). Primary care weaknesses were linked to health worker attitudes and, most importantly, the lack of a public family GP system that enables patients to return to the same doctor(s) of their choice and develop relationships of trust over longer periods of time. For the poorest households, a visit to a private doctor cost roughly the equivalent of a daily wage. Although people were willing to pay a private ‘family’ doctor for better relationships and patient focus, rising debt levels and asset depletion raised questions about the affordability of these payments (Russell 2001). The case study work therefore identified key dimensions of quality of care that needed improvement to increase access to and uptake of public health care benefits, and to expand the protection which the Sri Lankan health care system aims to achieve.

Thirdly, the intensive study of a small number of families revealed a range of assets and strategies important for coping with illness costs and increasing resilience to illness-induced impoverishment, notably social networks and formal and informal financial institutions that enabled saving and access to cheap credit. The study showed that the poorest households had weak social networks that were not a reliable source of financial support at times of illness, and they lacked access to financial institutions. The case study approach therefore highlighted a need for multi-pronged approaches, through government and NGO interventions beyond the health sector, to increase resilience to illness-related shocks and mitigate illness-induced impoverishment.

The knowledge claims of case study work are often criticized on the grounds that the evidence produced is ‘anecdotal’ or ‘unrepresentative’. But just as clinical science uses cases to understand disease causation, so social science can use cases to understand and build theory about poverty causation; it is one thing to identify vulnerable groups and their characteristics, and another to understand the mechanisms that have made them vulnerable and how these operate within households to ‘filter’ policy efforts and impact. Health policy researchers concerned with household processes must avoid what Inhorn (1995: 287) calls ‘methodolatry’, the adherence to a particular scientific methodology, in this case population-based analysis that relies on large samples, when alternatives can aid understanding.

The two strands of survey and case study work are complementary. Cases provide a powerful tool for understanding, providing answers to the how and why questions: why people have certain priorities and do or do not take up health care service benefits; why they choose private over public providers; why they use social grants in particular ways; how people access resources outside the home and how these resources increase resilience to illness costs; and how informal social protection mechanisms might be supported by interventions. They provide rich illustrations (rather than statistical representations) of social processes which can be hard to capture in a survey.

However, the policy relevance of case study and qualitative material does rely on it being ‘typical’ for a larger group of households (Coast et al. 2004), even though ‘atypical’ households can make revealing cases (Mitchell 1983). It therefore relies on the careful selection of cases from different population groups of relevance to the study and policy. Household surveys can create a sampling frame and establish categories from which to select cases, for example the asset rich or poor, those above or below a poverty line or who receive a social grant.

To date there have been few longitudinal studies on the economic burden of illness, the links between illness and impoverishment, and the factors mitigating impact. Most notably, scope remains for more longitudinal research on the economic burden of serious illnesses with long-term impacts such as TB, HIV/AIDS and a range of chronic illnesses increasingly prevalent in developing countries. This research is necessary to inform the development of health policy and health care delivery arrangements that increase access and reduce cost burdens. To take this field of research forward, this paper has argued for research designs that include a longitudinal case study component to enrich, and provide context-specific, understanding of the four economic burden indicators described at the start of this paper, and thus provide additional insights for health and social policy measures that aim to protect households from illness-induced impoverishment.

Biography

Steven Russell graduated from Cambridge University in 1988, lived and worked in Sri Lanka for several years and did a Masters in Development Studies at the University of East Anglia in 1992/3. He started working on health policy issues in 1993 when he joined the Health Policy Unit at the London School of Hygiene and Tropical Medicine, UK. His PhD was based on research in Sri Lanka on the economic burden of illness for households. Now a lecturer at the School of Development Studies at the University of East Anglia in the UK, his research interests cover a variety of health policy questions and are currently focused on the user- or demand-side issues of treatment behaviour, vulnerability and the economic burden of illness for households and the links between illness, livelihood security and poverty.

Thanks to Lucy Gilson and Anne Mills for their comments on earlier drafts of this paper. The paper was written with the financial support of the Health Economics and Financing Programme (HEFP) at the London School of Hygiene and Tropical Medicine, UK, funded by the UK Department for International Development (DFID). DFID supports policies, programmes and projects to promote international development and provided some of the funds for this study as part of that objective, but the views and opinions expressed are those of the author alone.

References