ARTICLE IN PRESS

0277-9536/$ - se

doi:10.1016/j.so

�CorrespondE-mail addr

lgilson@iafrica

Social Science & Medicine 63 (2006) 1732–1744

www.elsevier.com/locate/socscimed

Are health services protecting the livelihoods of the urban poorin Sri Lanka? Findings from two low-income areas of Colombo

Steven Russella,�, Lucy Gilsonb

aUniversity of East Anglia, Norwich, UKbLondon School of Hygiene and Tropical Medicine, UK

Available online 12 June 2006

Abstract

Investing in pro-poor health services is central to poverty reduction and achievement of the Millennium Development

Goals. As health care financing mechanisms have an important influence over access and treatment costs they are central

to the debates over health systems and their impact on poverty. This paper examines people’s utilisation of health care

services and illness cost burdens in a setting of free public provision, Sri Lanka. It assesses whether and how free health

care protected poor and vulnerable households from illness costs and illness-induced impoverishment, using data from a

cross-sectional survey (423 households) and longitudinal case study household research (16 households). The findings

inform policy debates about how to improve protection levels, including the contribution of free health care services to

poverty reduction. Assessment of policy options that can improve health system performance must start from a better

understanding of the demand-side influences over performance.

r 2006 Elsevier Ltd. All rights reserved.

Keywords: Poverty; Vulnerability; Illness cost; Coping strategy; Sri Lanka

Introduction

Illness can cause impoverishment through adownward spiral of income loss, treatment costsand asset depletion. Investing in pro-poor healthservices is therefore central to poverty reduction andachievement of the Millennium Development Goals(WHO, 2002; World Bank, 2004). As health carefinancing mechanisms have an important influenceover access and treatment costs they are central todebates over health systems and their impact onpoverty (Kawabata, Xu, & Carrin, 2002; World

e front matter r 2006 Elsevier Ltd. All rights reserved

cscimed.2006.04.017

ing author. Tel.: +44 1603 593373.

esses: s.russell@uea.ac.uk (S. Russell),

.com (L. Gilson).

Bank, 2004; Xu et al., 2003). Calls for the removalof user fees at public health care facilities have, thus,once again come to the international policy agenda(Commission for Africa, 2005). Out of pocketpayments add to the other barriers that poor peopleface when seeking health care, and contribute totheir experience of social exclusion. Even relativelysmall health care payments might push vulnerablehouseholds into absolute poverty or deepen theirpoverty (Gilson & McIntyre, 2005).

Discussion of financing policy options to improvehealth system performance in resource poor settingsmust, however, start from better understanding ofdemand-side factors and the consequences of out ofpocket payments for household poverty and liveli-hoods. This paper uses a household livelihood

.

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1733

framework to examine people’s utilisation of healthcare services, illness costs and their implications forimpoverishment in a setting of free public provision,Sri Lanka. It examines whether and how free healthcare protected poor and vulnerable householdsfrom illness costs and illness-induced impoverish-ment, informing national policy measures to im-prove protection levels and international debates onthe contribution of free health care services topoverty reduction.

Sri Lanka provides a particularly relevant casestudy with which to examine these issues. Histori-cally the country has been relatively successful in‘making services work for poor people’ (Rannan-Eliya, 2001), benefiting from the ‘long route’ togovernment and provider accountability to the poor(World Bank, 2004). Since democratisation in the1930s, competitive politics, left-wing political par-ties, trade unions and public pressure have con-structed a strong policy discourse that makes it thestate’s responsibility to deliver free health care as abasic right for all citizens (Sen, 1988). AfterIndependence government invested in a networkof accessible and free health care services and welltrained nurses and doctors. Effective use of thisnetwork by a well-educated population, notablyliterate women, helped bring about ‘good health atlow cost’ in Sri Lanka (Halstead, Walsh, & Warren,1985).

The health care market, however, has beenchanging, with private sector expansion since the1980s and a slow public sector response to changingdisease burdens and patient preferences. Detaileddemand analysis is therefore appropriate to assesscurrent patient utilisation patterns in this morecomplex market, as well as the levels of protection,and gaps in coverage, offered to poor households byfree public health services.

A household livelihood framework to inform pro-poor

health services

Study setting and the conceptual framework used in

the research

The research on illness and its livelihood impactwas conducted between 1998 and 1999 in two low-income settlements of Colombo, the capital of SriLanka. The urban sites were characterised byovercrowded housing, poor sanitation, drug abuseproblems and low incomes due to uncertain anddaily employment opportunities. These livelihoods

contributed to income poverty and vulnerability towage losses caused by incapacitating illness. Thesettlements lie a few miles from the centre ofColombo and close to many health care providers:a local municipal dispensary where a GP can beconsulted with no charge; several Ministry ofHealth tertiary hospitals where services are free tothe user; and a large number of private GPs,pharmacies and several private hospitals. Although,the research was conducted 8 years ago healthservice financing and delivery arrangements remainthe same at the time of writing. The two case studyurban areas were selected because they were typicalof many deprived settlements in Colombo.

The research objectives were to record treatmentseeking behaviour, measure the household costs ofillness, and assess coping strategies and theirconsequences for the household economy. Theconceptual framework that guided the research(see Russell, 2004) was based on inter-disciplinaryapproaches that have analysed the numerousresources people draw on to promote health orcope with illness costs (Berman, Kendall, &Bhattacharyya, 1994; Wallman & Baker, 1996) aswell as a livelihood framework (Scoones, 1998).Direct illness costs and indirect costs are defined,respectively, as expenditure linked with seekingtreatment and income losses caused by illness. Theterm ‘cost burden’ refers to direct or indirect costsexpressed as a percentage of household income.Health care spending and income losses will reducehousehold budgets and threaten members’ mini-mum basic needs such as food consumption oreducation, triggering coping strategies such asborrowing or asset sales. The resource strategiesused to cope with illness costs were also recordedbecause such strategies can mitigate or exacerbatethe overall economic impact of illness for thehousehold. Together illness costs and copingstrategies have implications for household income-poverty and livelihood outcomes, assessed usingindicators such as changes to income, working days,assets, consumption levels and food security.

Household vulnerability or resilience to illnesscosts is defined as the capacity to cope with illnesscosts without long-term damage to assets andimpoverishment. It is linked, first, to illness severity,with higher costs and less sustainable copingstrategies likely as severity and duration of illnessincrease. Second, capacity to cope is influencedby household asset portfolios (physical and finan-cial capital, human capital, social networks) and

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441734

policy-related resources that include health servicesas well as other public policy measures (e.g.education services) or community-based initiatives(e.g. micro-credit institutions) that contribute toresilience. These policy and community-based re-sources represent entry points for health and othersocial policy interventions that may protect house-holds.

Research methods

The research design had three phases spanning 18months. First, individual and group interviews wereconducted to generate qualitative data on treatmentbehaviour and livelihood difficulties. Second, across-sectional survey of 423 households and 2197individuals produced a statistical profile of house-hold income and assets, illness episodes, treatmentactions, illness costs and coping strategies. Thehouseholds were selected by systematic randomsampling and the sample covered 20% of the 2100households in both settlements.

The survey collected data on three categories ofillness expected to cause different treatment, costand coping patterns:

�

Acute illness episodes in the previous 2 weeks(except hospital admission). � Chronic illness in the previous month, cate-

gorised as such if the condition had persistedfor over 1 month or the respondent knew thediagnosis and the name of the chronic condition(e.g. diabetes, high blood pressure); the recallperiod allowed the survey to capture patients’regular monthly visits to providers.

� Hospital inpatient (IP) treatment in the previous

year, with the recall period designed to maximisehospitalisation events recorded.

The survey estimated household income usingdetailed consumption and expenditure questions.There is limited seasonality of casual labour or wagelevels in Colombo so expenditure or income levelswere not influenced by the timing of the survey. Inmost cases either the household head (usually male)or their partner (usually wife) was interviewed, andsometimes more than one adult was present. Wherepossible the wife or mother was asked questionsconcerning illness and treatment among familymembers.

All illness cost data were converted to a cost permonth figure to allow a total illness cost burden per

month to be calculated and analysis of the effect ofhealth care spending on the monthly householdbudget. Patient and caregiver days off work due toillness were converted to a lost income figure usingan average daily wage derived from the local setting(Rs. 150 or US$2.30 per day). Only days lost byeconomically active members were included in theindirect cost calculations because valuing unpaidactivities is both fraught with difficulties and lessimmediately relevant to understanding the econom-ic burden of illness.

The third phase of the research was an in-depthlongitudinal study of 16 case study households over8 months conducted to allow detailed investigationof illness costs and livelihood impacts over time.Using the survey data as a sampling frame thehouseholds were selected purposefully to be ‘typical’of four per capita income quartile groups and,within each, a range of illness, treatment and costexperiences. Finally, the selection process ensuredthat households with varying vulnerability orresilience to these costs were included within thecase studies. As assets, like income, reflect ability tocope with illness costs and livelihood change,assessment of household vulnerability or resiliencewas based on a simple audit of assets: the number ofworkers and security of work; physical capitalincluding house construction; education; and finan-cial capital.

Each household was visited at least every 2 weeks.Structured interviews were used for more quantita-tive variables (expenditure, illness costs, borrowing),and semi-structured interviews and observation togenerate qualitative data. The intensive study of asmall number of families over time was necessary toexplore the ways that people took action in theirevery day lives to treat illness and cope with itscosts, a well as the multiple factors that mediatedthe impact of illness on livelihood outcomes.

The knowledge claims from case studies are oftencriticised on the grounds that the evidence is‘anecdotal’ or ‘unrepresentative’. But just as clinicalscience uses cases to understand disease causation,so social science can use cases to understand illness-induced poverty causation. Such understandingmust go beyond the identification of vulnerablegroups’ characteristics to consider the social pro-cesses that cause vulnerability to illness costs andhow these operate within households to ‘filter’policy effects. As case study data are not statisticallyrepresentative but aim to strengthen understandingof social processes, sample size is of less concern

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1735

than the depth of understanding generated. Gen-eralisation is possible in terms of the concepts orframeworks (e.g. vulnerability) developed from casestudy analysis that can be applied to otherindividuals, households and settings (Coast, 1999;Mitchell, 1983).

The policy relevance of case study material does,however, rely on it being ‘typical’ for a larger groupof households, requiring the careful selection ofcases from different population groups of relevanceto the study. Here, use of the survey data enabledthe selection of cases that were typical of differenthousehold types in the two settlements, providingthe basis for the conceptual generalisation of theirexperiences to other households with similar char-acteristics in the same communities, such as theincome- or asset-rich and poor.

Illness costs and livelihood change: an overview

Cost burdens

Among the 323 households (out of 423) thatexperienced illness and self-treated or sought treat-ment, the median direct cost of illness was US$2.10(Rs. 138) per household per month or equivalent tojust under an average daily wage. The mean direct

0

10

20

30

40

50

60

70

80

0

0.1-

1.0

1.1-

2.0

2.1-

3.0

3.1-

4.0

4.1-

5.0

Cost burden (cost as

% o

f ho

use

ho

lds

(n=3

23)

direct cost burden indi

Fig. 1. Distribution of illness cost

cost was higher at US$7.50 (Rs. 487) per monthbecause a minority of households experienced ahigh direct cost. A mix of public and privateproviders was used (see Section ‘Protecting thepoor? Universal coverage and its limitations in SriLanka’). The main direct cost components fromprivate sector use were consultation fees andmedicine and the main cost item from public sectoruse was transport. No ‘under the table’ paymentswere recorded.

The majority of households (77% or 250/323)that experienced illness incurred a low or moderatedirect cost burden of 5% or less of monthly income(Fig. 1), either because the illness was mild orbecause free public services protected against highor catastrophic cost burdens associated with seriousillness. Low direct costs were not caused by peoplefailing to seek the medical care that they needed.However, a considerable minority of householdsexperienced what some analysts have called a‘catastrophic’ direct cost burden in terms of itspotential consequences for poverty (Prescott, 1999;Ranson, 2002): 10% of households (n ¼ 32) in-curred a direct cost burden above 10% of monthlyincome (Fig. 1).

The majority of households incurred no or lowindirect cost burdens (Fig. 1). Many illnesses did not

5.1-

6.0

6.1-

7.0

7.1-

8.0

8.1

9.0

9.1-

10.0

10.1

+

% of monthly income)

rect cost burden total cost burden

burdens across households.

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441736

cause income loss because children of school agedisproportionately suffered from acute illnesses, alarge proportion of acute illnesses experienced byeconomically active adults were not serious enoughto affect work, and the majority experiencingchronic illness and hospital admission were eco-nomically inactive. However, a minority (11%,n ¼ 35) incurred an indirect cost burden above10% of normal monthly income (Fig. 1).

Combined (total) cost burdens were relatively lowfor the majority of families surveyed (Fig. 1).However, a fifth (19.2%, n ¼ 62) incurred a totalcost burden above 10% and most of this groupincurred a total cost burden between 10.1% and40.0%.

Households in the poorest income quartile weredisproportionately affected by a catastrophic directcost burden above 10% because of their particularlylow income. However, there was no statisticallysignificant difference in mean direct cost burdensacross income groups (Table 1). Low median directcost burdens reflect the public health system’scoverage of the majority.

Case study households’ average direct (andindirect) cost burdens per month over 8 monthsare plotted in Fig. 2, with the households groupedinto the three vulnerability categories determinedfrom asset portfolios (see Section ‘Illness-relatedpoverty and livelihood change’). The majorityexperienced a low to moderate direct cost burdenper month under 5% of income, but a minority(n ¼ 3), all in the middle (vulnerable) groupexperienced a higher direct cost burden over 5%and one (Geetha) over 10%. Highly vulnerablehouseholds’ low average direct cost burdenstemmed from their greater use of free publicproviders. Resilient households’ low direct costburdens per month were because of their higherincomes and use of public hospitals for IP treatment(see Section ‘Protecting the poor? Universal cover-age and its limitations in Sri Lanka’).

For all case study households the average costburden per month conceals fluctuations over 8

Table 1

Average household direct illness cost burden per month by income qua

Direct cost burden Household income per capita qu

1 (poorest) (n ¼ 82) 2 (n ¼

Mean (95% confidence interval) 15.2 (�6.6–37.0) 3.0 (2

Median 1.2 1.4

months and usually in 1 or 2 months direct (andindirect) cost burdens were particularly high (seeFig. 3). This ‘lumpy’ feature of illness costs madethem harder to manage. Even the peaks in Fig. 3were average cost burdens over 30 days that smoothhigher daily cost burdens, often exceeding 100% ofthe daily wage. The poorest and most vulnerablehouseholds dependent on a low daily wage found itdifficult to manage any cost associated with illness,let alone these peaks, and had to borrow or pawnjewellery to cope.

A cost burden figure only indicates the potentialor likely consequences of an illness cost for house-hold impoverishment. The actual impact willdepend on household income (for a poor householda relatively low cost burden may cause impoverish-ment but for a non-poor household a burden above10% may not be ‘catastrophic’) and householdcapacity to mobilise additional resources (vulner-ability or resilience). However, it is still a usefulindicator of the extent of protection provided bypublic health services.

Illness-related poverty and livelihood change

The survey data were analysed to estimate theshort-term poverty implications of health carespending, using two indicators: the poverty count(incidence) and the poverty gap. The first calcula-tion estimates the proportion of households pushedbelow a US$30.00 per capita per month (US$1.00 aday) absolute poverty line by health care spending.Household health expenditure was subtracted fromhousehold income and a new household per capitaincome level calculated. As a result of health carepayments the poverty incidence rose from 54.1%(n ¼ 229) to 57.0% (n ¼ 241): 12 households werepushed below the poverty line. Health care spendingtherefore added 2.9% to the poverty incidence, alevel comparable to estimates from India andVietnam (Wagstaff, 2002). This analysis assumesthat the money spent on health care was no longeravailable to spend on other essential goods and

rtile

artile group

90) 3 (n ¼ 74) 4 (n ¼ 77) All (n ¼ 323)

.1–3.9) 3.5 (2.4–4.5) 4.1 (2.8–5.5) 6.5 (1.0–12.0)

1.8 1.6 1.3

ARTICLE IN PRESS

0 5 10 15 20 25 30 35

Sumithra

Valli

Jayasinghe $$

Selvaraja

Nimal $$

Geetha

Renuka

Amali

Kumudu

Nishanthi

Raja

Pushpa

Mayori

Rani

Mary

Dilani

Ho

use

ho

lds

Average total cost burden per month (% of income)

direct cost burden indirect cost burden

Resilient

Vulnerable

Highly vulnerable

Fig. 2. Average illness cost burden per month among case study households. $$Nimal and Jayasinghe were the main breadwinners in their

households but had been forced to stop work due to serious illness before research started. Indirect cost burdens were therefore high but

incalculable. Nimal experienced high direct costs of illness over 8 months but his extended family paid these costs.

S. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1737

services and so pushed households into absolutepoverty. It might also be argued that without freehealth care there would have been higher levels ofspending and the potential for more households tohave fallen below the poverty line.

The second calculation uses the poverty gapindicator (the average income shortfall from thepoverty line) to estimate the deepening of householdpoverty caused by health care spending. Among the229 households below the US30.00 poverty line themean income shortfall was US$8.90 (Rs. 577) percapita per month, or a daily shortfall of US$0.30below the US$1.00 a day poverty line (Table 2).After health care spending among the same house-holds the poverty gap rose to US$9.30 per month, a5.2% rise in the depth of poverty. If the 12additional households that fell below the povertyline are included in the calculation the depth ofpoverty rises by only 0.82%, from US$8.90 toUS$9.00 per capita per month (Table 2).

These indicators of changes to poverty derivedfrom a cross-sectional survey should be interpretedcautiously. A fall below the poverty line for examplemay be very short-term, households may havecoped by mobilising other resources, and the healthcare spending may not have involved any damagingcuts to consumption or assets. The advantage of thelongitudinal case study methodology was the abilityto track the actual implications of illness costs forincome-poverty, assets and livelihood change insome detail over 8 months.

The 16 households were chosen to represent fourhousehold income groups derived from the surveydata, but the other selection criteria (illness andvulnerability) meant they were not equally distrib-uted across quartiles (Table 3). Households in thelowest two quartiles earned less than US$1.00 percapita per day (less than US$30.00 per month). Inthe poorest quartile households struggled to meetfood and fuel needs on a daily basis. Even in the

ARTICLE IN PRESS

Table 2

Changes in the depth of household poverty (poverty gapa) due to health care payments

Mean poverty gap US$

(Rs.)bStandard deviation US$

(Rs.)

Poverty gap before health spending (n ¼ 229) 8.90 (577) 5.6 (367)

Poverty gap after health spending (n ¼ 227)c 9.30 (607) 5.6 (361)

Poverty gap including 12 new households below the

poverty line after health spending (n ¼ 239)

9.00 (582) 5.7 (370)

aAverage shortfall from a US$30.00 per capita per month poverty line.bAt the time of research US$1.00 ¼ Sri Lankan Rupees (Rs.) 65.00.cTwo outlier households excluded: very high health expenditure had pushed the households’ income into a negative income value that

prevented analysis of the poverty gap.

0

5

10

15

20

25

30

35

1 2 3 4 5 6 7 8

Month

Co

st b

urd

en p

er m

on

th (

% o

f in

com

e)

Pushpa - direct cost Selvaraja - direct cost

Raja - indirect cost Valli - indirect cost

Fig. 3. Monthly fluctuations of direct and indirect illness cost burdens: selected cases.

S. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441738

upper quartile most households earned onlyUS$40–50.00 per capita per month (US$1.00–2.00per capita per day). So despite their relatively highcash income in these poor areas many familiesclassified as ‘better-off’ were only marginally abovethe poverty line.

Seven of the 16 case study households were in thepoorest quartile and an additional expense such ashealth care usually triggered coping strategies thatpushed them deeper into poverty. The two house-holds in the second quartile also had little money

available for health care. Income insecurity due tolack of available work or illness was a great sourceof vulnerability:

Illness is something we are all scared of here.How can we live without working? If myhusband is ill we have to get money fromsomewhere for food and for the medicine, wehave to borrow.(Selvaraja, woman from poorest income quartile,most vulnerable).

ARTICLE IN PRESS

Table 3

Location of case study households in the community income profilea

Household income per capita quartile group

1 2 3 4

Household per capita income: US$/month

(Rs./month)

$ 0–21

(0–1352)

$ 21–29

(1353–1880)

$ 29–40

(1881–2609)

$ 40+

(2610+)

Households with illness Nimal Nishanthi Raja Dilani

Jayasinghe Sumithra Rani

Valli

Households without illness Kumudu Pushpa Mary

Selvaraja Mayori Renuka

Amali

Geetha

aCase study households have been given pseudonyms for confidentiality.

S. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1739

Households in the third and fourth quartilescould to differing degrees meet the costs oftreatment for most acute and chronic illnesses inmonths when incomes were maximised. However, inmonths when workers lost earnings due to illness orthe vagaries of the labour market, or when illnessexpenses coincided with other ‘lumpy’ expenses suchas education or clothing, those in the third quartilehad to adopt strategies to cope with illness costs.Those in the fourth quartile had to mobiliseadditional resources when more serious or pro-longed illness caused income loss. In other wordshousehold ability to cope with illness costs couldnot be seen in isolation from other expenses andincome fluctuations.

Across income groups, case study householdswere also selected from three vulnerability—resi-lience categories (Fig. 2). Over 8 months, livelihoodchange among the households was evaluated byanalysing six livelihood outcomes using quantitativeand qualitative data: the number of workers and jobsecurity; income levels; physical capital; financialcapital (changes to savings or jewellery); debt levels;and consumption (focusing on number of meals perday). Households were placed into three categoriesof livelihood change: struggling (impoverishment);coping (stability); investing (improvement).

Highly vulnerable households: struggled and became

more impoverished

Three out of four households in this group (seeFig. 2 for pseudonyms) were located in the poorestincome quartile (Table 3) and struggled to eat threemeals a day. They had weak asset portfolios.Members had less formal education and relied on

one or sometimes two workers with insecure jobs.Physical capital was limited to a small wooden orpoorly maintained cement block house with noelectricity or water connection. Financial capitalhad been depleted: they had pawned all or most oftheir jewellery, in some cases due to previous illness(Jayasinghe, Valli), and were in debt to money-lenders.

Over the 8 months these households experienceda decline in at least four of the six livelihoodvariables, most commonly the loss of an incomeearner or growing insecurity of work, pawnedjewellery, increased debt and lasting cuts to foodconsumption. Three of the four households were ona path of decline triggered by illness before researchstarted. For example cancer had forced Jayasingheto give up work with damaging economic conse-quences for the household; and Sumithra’s husbandhad experienced a serious accident which, after overa month in (a public) hospital without earningincome, had undermined assets and caused highlevels of debt.

Three of the highly vulnerable households in-curred low or moderate average cost burdens permonth (Fig. 2) but these costs were a persistentattack on the household budget and assets. Valliexperienced indirect cost burdens of over 20% insome months (Fig. 3) which could be judged to be‘catastrophic’ because they forced her and herhusband deeper into poverty: they had to borrowat high interest, cut food consumption and pawnlast items of jewellery. The group’s low and insecureincomes meant they had to meet a high propor-tion (58%) of direct and indirect illness coststhrough these types of strategy, but their weak

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441740

asset portfolios meant they struggled to cope.Multiple asset weaknesses made health serviceprotection particularly important for this group ofhouseholds (see Section ‘Protecting the poor? Uni-versal coverage and its limitations in Sri Lanka’).

Vulnerable households: coped to different degrees

This group spanned the full range of per capitaincome quartiles and to differing degrees hadstronger asset portfolios than the highly vulnerablegroup, even among the income-poorest (Amali,Nimal, Geetha). Renuka’s household was located inthe top income quartile but was vulnerable becauseher husband used the income to fund a heroinaddiction and the rest of the family (Renuka andfour children) were left with barely enough incomefor food and few assets. Compared to the highlyvulnerable group, adults were in general bettereducated (Nimal, Geetha, Amali, Pushpa) or thehousehold had more workers (Raja, Pushpa,Nishanthi). Some had more financial capital withwomen participating in rotating savings (seetu)groups or credit societies (Nishanthi, Amali, Ku-mudu, Raja, Pushpa), although Nimal’s wife Sita,Geetha, and Renuka were not involved due to theirincome-poverty. Some had jewellery available topawn (Geetha, Amali, Kumudu, Raja, Pushpa), butothers had depleted these financial assets due toprevious illness (Nimal, Nishanthi).

These households experienced little change to atleast four dimensions of livelihood. Debt levels hadnot increased and if people had borrowed it wasfrom low cost and flexible sources such as family,friends and local credit societies. Historically theywere on steady livelihood paths characterised byvulnerability but fewer shocks including fewerserious illness events. Gradual improvements weresustained by Kumudu and Puspha despite highillness costs (Fig. 2) and the others were coping todiffering degrees. As Nimal’s household had alreadysuffered dramatic decline due to serious illnessbefore the start of the study, it could have beenplaced in the struggling group. However, whenresearch started he and his wife were coping (at alower level) and not suffering further impoverish-ment because free health services enabled him tomake regular visits to the hospital for consultationsand blood tests, and strong family networksprovided funds for nearly all their daily and healthcare expenses.

Despite the group’s higher direct cost burdens,over 100% in some months for Geetha and Nimal,

this group was distinguished from the highlyvulnerable households by their stronger assetportfolios and capacity to cope, particularly thestrength of their social networks. Although incomepoverty meant the group could not cover aconsiderable proportion (45%) of their total illnesscosts through usual income sources, they mobilisedlow cost asset and borrowing strategies, whichcontributed to livelihood stability. Access to freeservices for more serious illness contributed to thisresilience (see Section ‘Protecting the poor? Uni-versal coverage and its limitations in Sri Lanka’).

Resilient households: invested and improved

These four households had higher and moresecure incomes derived from a household memberwith a secure government job, or several workers inthe family, or a successful small business. They hadthe strongest asset portfolios including bettereducation, a larger house made from bricks andmortar, and a range of physical assets in thehousehold (electrical goods, furniture).

Over 8 months the group experienced improve-ment in at least four livelihood variables, and nearlyall borrowing was for investment purposes. Histori-cally they were on steady trajectories of improve-ment even though they had originally started fromsocio-economic positions similar to the other house-holds. Notably no breadwinners had been affectedby serious illness.

Although household members used private pro-viders more often than public providers for treat-ment of acute and chronic illnesses, the groupexperienced relatively low or moderate cost burdensbecause of their higher income (Fig. 2). However,they relied on the safety net or ‘insurance’ of thepublic sector for IP treatment, which protectedassets, kept debts low and also allowed them todivert resources to investment strategies. As a result,they only had to cover a small proportion (9%) oftotal illness costs through asset strategies, usuallylow cost borrowing from strong social networks.

Across these three groups of household there was,not surprisingly, a strong link between vulnerabilityat the start of research and livelihood changecategory at the end. Fig. 2 also suggests there wasno clear link between illness cost burden andlivelihood change. Highly vulnerable householdswith direct cost burdens less than 5% struggled andfell further into poverty. In contrast some of themiddle (vulnerable) group incurred a high or ‘cata-strophic’ burden but managed to cope, although

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1741

households experiencing serious illness and a highcost burden (Geetha and the special case of Nimal)were only just coping. Given the complexity oflivelihoods and the multiple factors influencinglivelihood trajectories, the lack of a clear linkbetween cost burden and impoverishment is notsurprising.

Protecting the poor? Universal coverage and its

limitations in Sri Lanka

Inpatient treatment

The household survey found that the vastmajority of people in the two communities, fromall income groups, used one of the large publichospitals in the city rather than a private hospital(98% of admissions, n ¼ 177). Among case studyhouseholds all hospital admissions over the 8-month period were to public hospitals. This utilisa-tion pattern was explained by the free IP careoffered by public hospitals compared to theprohibitively high cost of a private hospital admis-sion, but in addition a dominant theme from thequalitative data was people’s trust in the technicalquality of care at public hospitals, based on thewidely held view that they had the best staff andequipment to deal with serious conditions (Russell,2005).

Use of public hospitals meant patients and theirfamilies incurred a relatively low direct cost burdenfor a hospital admission. From the householdsurvey 82% of households experiencing one ormore IP admission (n ¼ 134) in the previous yearfaced a direct cost burden of 1% or less per month(i.e. a burden of 12% or less in 1 month spread over12 months), and 95% of households experienced aburden of 5% per month or less, the main cost itemsbeing transport, special food and complementaryreligious therapies or medicine bought outside thehospital. Case study household experiences alsodemonstrated how free IP services protected allsocio-economic groups against high direct medicalcosts.

Regular treatment of chronic illness

Chronic conditions requiring regular treatmenthave the potential to impose high- and long-termcost burdens on poor households unless free servicesare available. The household survey identified 342people (15.6%) who reported a chronic condition

and 194 sought treatment on a regular basis. Acrossthe first three income quartile groups a publichospital OP clinic was the main source of regulartreatment, particularly among patients from thepoorest households who used public providers farmore frequently (62%) than private providers(18%). Only the ‘better-off’ income group usedprivate providers more often than public providersfor their regular treatment. Free treatment was themost common reason cited for using a publicprovider but confidence in the technical competenceof doctors was again important (Russell, 2005).Widespread use of public providers meant that outof the 155 households with a member seekingregular treatment 50% incurred a direct cost burdenof 1% of monthly income or less, 87% a burden of5% or less and only 3% of households incurredregular monthly burdens over 10%.

The case study data confirmed that free healthcare offered important protection to livelihoods.For the highly vulnerable group with no surplusmoney to pay for health care (even to covertransport costs), free regular treatment of chronicconditions was vital protection against higherborrowing or deeper cuts to food consumption.Among the vulnerable (middle) group free treat-ment was also a vital entitlement that preventedborrowing for health care expenses. A comment byGeetha, diagnosed with Type 2 diabetes during the8-month study period, exemplifies the experience ofdiabetics from vulnerable households:

If I go private, I pay money, but then if things getworse they refer me to the government and theywould have to do all the tests again. So if I have abig problem, or one that needs continuoustreatment like diabetes, I go to the governmenthospitalyIt is freeyhow could I pay for thetablets everyday?(Geetha: woman from poorest income quartile,vulnerable).

Free treatment was particularly important tolivelihood security at times when workers fell illcausing income levels to drop and a consequentstruggle to pay for a range of essential items. Raja’shousehold, for example, experienced high wagelosses in some months (see Fig. 3) because he andhis wife (Ranji) suffered from asthma. In month 1Raja had a sore chest and took two days off work,losing US$5.40 (Rs. 350) in wages (an indirectcost burden of 6%). Raja went to a nearby privateclinic and pharmacy for treatment, which incurred a

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441742

direct cost burden of 15%. The high direct costburden combined with the indirect cost forced thefamily to borrow from Raja’s workplace. Later inthe month the chest problems persisted but they hadno cash available so Raja resorted to the freemunicipal dispensary. Without the alternative ofcheap public treatment towards the end of themonth the household’s borrowing would have beensignificantly higher.

Acute illnesses requiring OP treatment

The survey identified 266 out of 2197 individuals(12.1%) who reported an acute illness episode in theprevious 2 weeks, the most frequent being cold,cough, fever, flu, headache, injury and diarrhoea.Self-treatment at home was the most frequentlyreported first response (58%), reflecting the mildnature of many of the illnesses.

In contrast to the dominant use of publicproviders for IP and regular chronic care, the useof health care providers outside the home was moreequally split between public and private providersfor moderate acute illnesses, with private GPs andpharmacies slightly more dominant. Even amongthe poorest quartile a considerable minority ofpatients (46%) used private doctors and pharmacies(Russell, 2005).

Widespread use of private providers meant higherhousehold cost burdens for OP treatment of acuteillness. Out of the 210 households experiencing oneor more acute illness episode, 47% experienced adirect cost burden of 1% or less but 20%experienced a burden over 5 and 7% a burdenabove 10%.

All case study household respondents, whethermale, female, poor or better-off, stated that theypreferred to use a private doctor or pharmacy forcommon illnesses. Income levels and cash avail-ability, however, influenced actual utilisation pat-terns. Members of the seven households in the topincome quartiles (with the exception of Renuka)consistently used a private GP with whom they werefamiliar (their ‘family doctor’). In the seven pooresthouseholds wage-earners used private doctors andpharmacies more frequently than public providersto obtain treatment quickly and avoid wage losses,but members who did not work used a municipaldispensary as frequently as private doctors andpharmacies.

The research identified several reasons for thelower uptake of free public health services for

common acute illnesses (Russell, 2005), includinglimited opening hours, long waiting times, shortconsultations and poor inter-personal quality. As aresult the majority of the ‘better-off’ and even aconsiderable minority of the poorest were willing topay to get quicker care, secure a longer consultationwith more patient focus, and build a long-termdoctor–patient relationship with a ‘family’ doctor.The poorest pawned jewellery and borrowed moneyto finance private treatment.

Nonetheless, free public health care of adequatequality offered important protection to the mostvulnerable and income-poorest households withseveral small children who experienced frequentand concurrent acute illness events. Selvaraja’sfamily offers a typical illustration of this protection.In month 5 the three children suffered illnessconcurrently (high fever and vomiting) and Selver-aja took them to the National Children’s HospitalOP department, a visit which incurred a direct costburden of only 0.5% (US$0.50/Rs. 30 for trans-port):

I take the kids to Lady RidgewayyRs. 300 ormore would have gone if I had gone priva-teyand I would need to borrow even moremoney for that—maybe with interest.(Selvaraja, woman from poorest income quartile,highly vulnerable).

The livelihood implications of having to pay forhealth care were starkly illustrated in the samemonth. The family spent an additional Rs. 320(US$5.00) on health care due to private sector useby Selvaraja’s husband (for a recurring shoulderinjury; he could not afford to miss work) andSelvaraja’s mother (for a tooth extraction; there wasa long waiting list at the public hospital). Theseprivate visits imposed a direct cost burden of 5%which exceeded the household budget after foodpurchases and triggered coping strategies thatpushed the household deeper into poverty. Theyhad to borrow from an ex-employer (Rs. 500), delaypayment of the electricity bill, delay debt repaymentto the local food shop, and could not redeem a ringthat Selvaraja had pawned in an earlier month topay for health care. If Selvaraja had taken herchildren to a private doctor that month the overalldirect cost burden for the family would have beenover 12%, forcing even more risky borrowing orasset strategies.

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–1744 1743

Discussion

The household survey and case study data showthat free health care services in urban Sri Lanka,financed through taxation, protected the majority ofpoor households against high out of pocket pay-ments for treatment at the time of illness. Thisprotection against even relatively low fees was animportant poverty reduction measure because, asshown by the case study findings, even a small directcost could cause impoverishment. Nine case studyhouseholds in the two poorest income quartiles(Table 3), selected to be typical of 50% of totalhouseholds in these settlements, relied on low paidand insecure work and struggled on less thanUS$1.00 per capita per day. These households hadlittle or no ‘ability to pay’ for health care aftermeeting basic food, shelter and fuel needs. Otheressential but ‘lumpy’ expenses, on education, rites ofpassage, housing or clothing for example, werealready beyond the household budget. Any healthcare expense, even a moderate direct cost burden of2.5–5% of monthly income, or a loss of income dueto illness, inevitably triggered borrowing, pawning,or cuts to food and education. Longitudinal researchshowed that when a low or moderate direct costburden affected a poor household only once or twiceover 8 months then recovery was easier and illnessmade little difference to poverty. However, frequentmoderate illness costs experienced by poor familieswith small children or a chronically sick memberwere a persistent attack on already overstretchedbudgets that contributed to debt accumulation, assetdepletion and made the household vulnerable toother shocks. Vulnerability to income losses causedby illness, as well as transport costs, increased theimportance of the protection against medical costsoffered by free health care services.

The case study data also showed that the relation-ship between cost burden and livelihood change iscomplex. Highly vulnerable households that experi-enced low or moderate burdens declined, but lessvulnerable households that experienced a high or‘catastrophic’ burden coped and remained stable.The longitudinal case study research could explorethe processes explaining the links between illness costand livelihood change, through retrospective inter-views (life histories) and the prospective 8-monthstudy. Multiple factors affected livelihood andpoverty trajectories over time, including problemsarising from legal expenses, drink and other drugproblems, earlier shocks, the loss of land or an

illness, or broken relationships. Previous events andprocesses had placed households on longer-termtrajectories of struggling, coping or improving, andpath dependency continued to influence livelihoodchange over the brief research period. Given thestrength of these trajectories the impact of illness onimpoverishment and livelihood was heavily depen-dent on its severity, frequency and duration. Lowand infrequent illness costs made little impact. Lowor moderate but more frequent illness costs exacer-bated vulnerability and livelihood decline. Seriousillness that caused high or catastrophic and persistentcost burdens could have a major negative impact onlivelihood paths. In Sri Lanka the availability of freepublic health services meant it was the indirect costsarising from serious illness, rather than direct costs,which were the most obvious cause of illness-inducedpoverty, as the examples of Nimal and Jayasinghedemonstrated.

Other studies have shown that the Sri Lankanpublic health system has a pro-poor benefit in-cidence and is among the most equitable in Asia(Rannan-Eliya & EQUITAP partners, 2005). Fromthe data presented here, free health care’s contribu-tion to protecting against illness-induced impover-ishment for the three household livelihood groupscan be summarised as:

�

Free treatment mitigated further impoverishment

of declining households: Already on trajectories oflivelihood decline, free health services mitigateddeepening poverty from illness by reducing directcost burdens. Low and insecure incomes, assetweaknesses and burdens imposed by otherexpenses, however, meant that free health careservices alone were not enough to preventlivelihood decline. This demonstrates the vitalimportance of free care for this group, and theneed for other interventions to build resilienceagainst illness costs and other shocks.

� Free treatment prevented the decline or impover-

ishment of relatively stable households: Freetreatment, particularly free IP treatment andregular treatment of chronic illnesses, preventedhigh cost burdens and contributed greatly tolower debts and the prevention of asset depletionamong this group. Free treatment was particu-larly important when income earners could notwork or at times when the household facedcombined expenses. By protecting assets freetreatment also made these households moreresilient to future shocks.

ARTICLE IN PRESSS. Russell, L. Gilson / Social Science & Medicine 63 (2006) 1732–17441744

�

Free treatment enabled investment by improving

households: Free hospital IP care acted as‘insurance’ that allowed households to allocateresources to saving and investment strategies,rather than having to save to finance the costs ofa future hospital admission.

Free IP hospital services meant the health systemprotected the full range of socio-economic groupscovered by the study. It also demonstrated thateffective protection requires a broad package ofcurative treatment that is free at the point ofdelivery. However, the findings also showed thatfree public services only protected the poor whenthey were of a quality acceptable enough to be used.Public health care services were less successful inprotecting patients against the direct costs of acuteillness requiring treatment outside the home becausepeople across income groups, even from the poorestincome quartile, preferred to use private providers(Russell, 2005).

Overall, these findings can be applied to similarurban settings in Colombo because the study sitesand households were selected to be typical of suchsettings and populations. In rural Sri Lanka incomepoverty is wider and deeper, and the direct costs ofillness likely to be higher due to transport costs.Protection against medical costs is therefore likely tobe even more important for poverty reduction andlivelihood sustainability in rural areas of the country.Although Sri Lanka’s universal provision modelfaces financing and quality problems, the govern-ment should not start charging the user to raiserevenue. Fees would undermine livelihoods and oneof the few pro-poor health systems in the world.

Acknowledgements

Thanks to Anne Mills for her comments. Thepaper was written with financial support from theHealth Economics and Financing Programme,London School of Hygiene and Tropical Medicine,UK, funded by the UK Department for Interna-tional Development. The views and opinionsexpressed are those of the authors alone.

References

Berman, P., Kendall, C., & Bhattacharyya, K. (1994). The

household production of health: Integrating social science

perspectives on micro-level health determinants. Social

Science and Medicine, 38(2), 205–215.

Coast, J. (1999). The appropriate use of qualitative methods in

health economics. Health Economics, 8, 345–353.

Commission for Africa. (2005). Our common interest. London:

Report of the Commission for Africa.

Gilson, L., & McIntyre, D. (2005). Removing user fees for

primary care in Africa: The need for careful action. British

Medical Journal, 331, 762–765.

Halstead, S. B., Walsh, J. A., & Warren, K. S. (1985). Good health

at low cost. New York: Rockefeller Foundation.

Kawabata, K., Xu, K., & Carrin, G. (2002). Preventing

impoverishment through protection against catastrophic

health expenditure. Bulletin of the World Health Organization,

53, 61–67.

Mitchell, J. C. (1983). Case and situation analysis. Sociological

Review, 31, 187–211.

Prescott, N. (1999). Coping with catastrophic health shocks. In

Paper presented at a conference on social protection and

poverty, Inter-American Development Bank, Washington,

DC.

Rannan-Eliya, R. P. (2001). Strategies for improving the health

of the poor: The Sri Lankan experience. In Paper prepared for

Health Systems Resource Centre, Department of Interna-

tional Development, UK, by the Institute of Policy Studies,

Colombo, Sri Lanka.

Rannan-Eliya, R. P., & EQUITAP partners. (2005). Compara-

tive experiences of tax-funded health care systems in Asia:

Bangladesh, Hong Kong SAR, India, Indonesia, Malaysia

and Sri Lanka. Equity in Asia-Pacific Health Systems:

Working Paper # 7.

Ranson, K. (2002). Reduction of catastrophic health care

expenditures by a community-based health insurance

scheme in Gujurat, India: Current experiences and

challenges. Bulletin of the World Health Organization, 80(8),

613–621.

Russell, S. (2004). The economic burden of illness for house-

holds in developing countries: A review of studies focus-

ing on malaria, tuberculosis and HIV/AIDS. American

Journal of Tropical Medicine and Hygiene, 71(Suppl. 2),

147–155.

Russell, S. (2005). Treatment seeking behaviour in urban Sri

Lanka: Trusting the state, trusting private providers. Social

Science and Medicine, 61, 1396–1407.

Scoones, I. (1998). Sustainable rural livelihoods: A framework for

analysis. IDS Working Paper 72, Institute of Development

Studies, Brighton.

Sen, A. (1988). Sri Lanka’s achievements: How and when. In T.

Srinivasan, & P. Bardhan (Eds.), Rural poverty in South Asia.

New York: Columbia University Press.

Wagstaff, A. (2002). Poverty and health sector inequalities.

Bulletin of the World Health Organization, 80, 97–105.

Wallman, S., & Baker, M. (1996). Which resources pay for

treatment? A model for estimating the informal economy of

health. Social Science and Medicine, 42(5), 671–679.

WHO. (2002). Improving health outcomes of the poor: The report

of working group 5 of the commission on macroeconomics and

health. Geneva: World Health Organization.

World Bank. (2004). World development report: Making services

work for poor people. Washington, DC: World Bank.

Xu, K., Evans, D. B., Kawabata, K., Zeramdini, R., Klavus, J.,

& Murray, C. J. L. (2003). Household catastrophic health

expenditure: A multi-country analysis. The Lancet, 362,

111–117.