Clinical Epidemiology & Public Health

1. Introduction

There is a version of this story that has, by now, become almost too familiar to hear properly: farmland at the edge of a South Asian city gives way to something improvised, the improvised settlement hardens into something semi-permanent, and public services — water, sanitation, health care — arrive later, if at all. Bangladesh has lived this pattern with unusual intensity, and Dhaka in particular is projected to become one of the most populous urban agglomerations anywhere within a generation, much of that growth absorbed not by planned housing but by slums (Adams et al., 2020). It would be convenient to treat this purely as a housing problem, and for a long time policy discourse more or less did. But the sociological literature on slum formation makes fairly clear that this framing understates things: as Ezeh et al. (2017) argue in their wide-ranging review, slums generate a distinctive, cumulative set of health risks that housing policy alone cannot resolve. It is, in other words, just as much a health systems problem.

Part of the difficulty is that "slum health" is not one problem; it is a bundle of quite different problems that happen to share a geography. Maternal and reproductive health has been one recurring thread — Hossain and Hoque (2005), studying delivery-care choices in Dhaka's slums, found patterns shaped as much by household economics as by clinical need, and comparable inequities in reproductive and child health services have been documented in Chandigarh (Gupta et al., 2008). Khan et al. (2012) push further, arguing — convincingly, I think — that treating "the urban slum" as a single analytic category obscures more than it reveals, since maternal health conditions varied meaningfully even between two slum populations in the same city. Hulton et al. (2007) offer a useful corrective, proposing a quality-of-care framework flexible enough to capture that variation. None of this is directly about chronic illness, but it establishes that slum health cannot be treated as a single, homogeneous category.

A second thread, closer to this paper's actual focus, concerns non-communicable and chronic disease. Zaman et al. (2015), drawing on Bangladesh's STEPS survey, documented substantial clustering of NCD risk factors among Bangladeshi adults generally, and slum-specific work has since sharpened that picture: Rawal et al. (2017) found meaningful NCD and diabetes burden among adults in Dhaka's slums specifically, while Mondal et al. (2019) reported comparably concerning behavioral risk-factor prevalence in a separate slum sample. Taken together, these findings suggest chronic disease in Bangladesh's slums is already substantial, and that care infrastructure has failed to keep pace.

Primary health care is, in principle, the mechanism through which health systems are meant to reach populations like these. Dodd et al. (2019), reviewing primary care organization across low- and middle-income Asia-Pacific countries, identified five areas where guidance remains thin: non-physician workforce development, NCD prevention within basic care, managerial capacity, community engagement, and — most relevant here — modernized health information systems. Saif-Ur-Rahman et al. (2019) reach a related conclusion from a different angle, mapping persistent gaps in primary care governance across low- and middle-income settings more broadly. These gaps map onto conditions in Bangladesh's slums, where record-keeping, to the extent it exists, rarely follows a patient from one visit to the next.

That continuity problem deserves emphasis, because chronic disease management is, almost by definition, longitudinal. Baum et al. (2017), drawing on a five-year realist case study in South Australia, distinguish "comprehensive" from "selective" primary care models, and the distinction is not merely academic: selective, narrowly vertical programs can show results for a single condition while still failing at the sustained, whole-person follow-up chronic illness actually requires. Slum health systems, out of necessity as much as design, often end up looking selective even where the epidemiology plainly calls for something more comprehensive.

Vulnerable urban populations are not, in any case, a single undifferentiated group, however convenient that assumption might be. Hjern et al. (2019), examining health-screening practices among migrant children in Europe, found considerable inconsistency in which protocols were applied to whom — a reminder that even well-resourced systems can fail to standardize care for populations that move, or arrive undocumented, or fall outside routine catchment planning. A related, less obvious failure mode appears in Gemaque et al.'s (2014) study of hospitalized infectious-disease patients in northern Brazil, where oral lesions — a real but secondary condition — went comparatively under-examined amid the pressure of managing the primary illness. The settings differ enormously, but the dynamic feels familiar: in strained systems, whatever isn't the immediate emergency gets deprioritized.

Some more encouraging work has come from efforts that deliberately build bridges across these gaps rather than assuming patients will find their own way in. The IMPACT initiative described by Russell et al. (2019) links researchers, clinicians, and vulnerable-community members across Local Innovation Partnerships in Australia and Canada, explicitly aiming to improve access for populations conventional models tend to miss. Satherley et al. (2019), evaluating the Evelina London Model of Care for children and young people, describe a broadly similar partnership logic applied elsewhere. Whether either model would translate to Bangladesh's very different context is an open question, but the underlying principle — that access problems are often solved locally, through partnership — seems plausibly transferable.

Self-management approaches offer a complementary, and perhaps under-appreciated, thread. The lay-led model tested at Stanford in the 1980s (Lorig et al., 1986) found peer-taught arthritis self-management produced outcomes comparable to professionally led courses, a finding substantial enough to be generalized into the Chronic Disease Self-Management Program and tested again across a broader range of conditions (Lorig et al., 1999). Its evolution into England's NHS-based Expert Patients Programme (Halligan et al., 2006) suggests something worth taking seriously: peer support, structured properly, can extend a health system's reach without proportional growth in clinical staffing — attractive, at least on paper, for a resource-constrained system like Bangladesh's, though we are unaware of it being formally piloted in a slum setting there.

None of this happens in a political vacuum. Greer's (2008) analysis of a critical juncture in European Union health service policy usefully reminds us that reform tends to cluster around windows of political opportunity rather than unfold gradually — a pattern this paper's policy discussion returns to.

It is against this backdrop — a genuinely heterogeneous slum population, a documented and growing chronic-disease burden, thin primary-care information infrastructure, and a scattered but suggestive literature on what has worked elsewhere — that the present paper positions itself, deliberately, as a secondary analysis rather than a fresh data-collection effort. Its empirical foundation is the cross-sectional survey of adult slum dwellers in Dhaka and Tongi conducted by Adams et al. (2020). Rather than repeat that fieldwork, we revisit its descriptive findings, place them alongside the literature summarized above, and use that synthesis to sketch a policy and information-systems framework aimed at strengthening chronic illness follow-up for this population. Whether the framework is right in every particular matters, frankly, rather less than whether it prompts a more serious conversation about what continuity of care could look like where, at present, it largely does not exist.

2. Methodology

2.1 Study Design

This paper is a secondary, narrative analysis. No new human-subjects’ data were collected for this work, and no direct contact with study participants occurred. The empirical material discussed here — sample characteristics, marital status distributions, and Poverty Probability Index (PPI) scores — is drawn entirely from the previously published cross-sectional survey of Adams et al. (2020), conducted among adult slum dwellers in two Bangladeshi urban sites. We describe that original design in reasonable detail below; both so readers can judge the strength of the underlying evidence and because — this is worth stating plainly — a secondary analysis is only as trustworthy as its ability to be traced back to a reproducible primary source.

2.2 Original Study Setting and Population

The source survey was fielded between 2013 and 2014 across two sites chosen, deliberately, for their contrasting health and economic profiles: Dhaka City Corporation, where roughly half the population lives in slum settlements sustained largely by day labor and small-scale trade, and Tongi, a township of approximately 400,000 residents within Gazipur City Corporation and a hub of the country's export-oriented ready-made garment and pharmaceutical industries (Adams et al., 2020). The contrast matters analytically — a labor-day economy and a factory-shift economy plausibly produce different illness patterns, different time constraints on care-seeking, and different exposure to workplace hazards, even within the broadly similar category of "urban slum."

2.3 Sampling Strategy

Community health workers in the original study were organized into geographic zones of roughly 200 households each, with each zone serving as a sampling cluster. In Dhaka, 30 of 189 available clusters were randomly selected, at a rate of 10 per zone; in Tongi, 34 of 330 clusters were selected using the same logic (Adams et al., 2020). Within selected clusters, households were visited directly and household members' age, sex, and recent illness experience were recorded, producing a stratified sampling frame by age and sex weighted against existing population health statistics. For the purposes of the present analysis, we retain the original study's focus on working-age adults, defined as those between 15 and 64 years, who reported a chronic illness lasting three months or longer.

It is worth noting — and Adams et al. (2020) are appropriately candid about this — that because sampling proceeded at the household level rather than the individual level, more than one household member could be interviewed if multiple residents fell within the eligible age and sex strata. This is a defensible design choice for capturing intra-household variation, though it does mean that observations are not strictly independent in a statistical sense, a point we return to in the Limitations section.

2.4 Instrumentation and Data Collection

The original survey instrument was designed to align with existing health-seeking behavior items validated in the Bangladesh Urban Health Survey and the Household Income and Expenditure Survey, which were administered entirely in Bangla (Adams et al., 2020). Four field teams, each comprising eight research assistants supervised by a researcher with prior survey experience, conducted data collection following a seven-day training period. A two-day pilot involving ten respondents in a Dhaka slum preceded full fieldwork, surfacing — as pilots usefully do — practical problems with the instrument: skipped items, ambiguous phrasing, and questions that were, on reflection, simply misread by respondents unfamiliar with survey conventions. The finalized questionnaire was administered by tablet, with supervisors reviewing entries in real time for inconsistency, and a central data-management team conducting a further consistency check before analysis.

2.5 Measures Used in This Analysis

For the purposes of this secondary synthesis, we focus on three measure sets reported in the original study and reproduced here in Tables 1 and 2: (a) sex and marital status distribution among working-age respondents at each site, (b) mean PPI score by site, a standard proxy for household poverty likelihood, and (c) PPI quintile distribution by site. These were selected because they are the variables most directly relevant to characterizing the vulnerability profile of each site's population — the analytic anchor for the policy discussion that follows in Section 5.

2.6 Analytic Approach

Our contribution at the analytic level is interpretive rather than statistical: we compare the reported site-level descriptive figures against the primary-care and chronic-disease literature reviewed in the Introduction, looking for points of convergence or tension. We did not conduct inferential statistical testing on the underlying data, and readers should not interpret the comparisons offered in Section 3 as hypothesis-tested findings; they are descriptive contrasts intended to motivate the policy framework proposed in Section 5. Any reader wishing to conduct further inferential analysis is directed to Adams et al. (2020), whose full dataset and methodology are described in that source publication.

2.7 Reproducibility Statement

Because this is a secondary analysis, reproducibility here means something specific: the descriptive figures reported in Tables 1 and 2 can be independently verified against Adams et al. (2020, PLOS ONE, https://doi.org/10.1371/journal.pone.0233635), and the sampling and instrumentation procedures summarized above are described in full in that source. We have not altered, re-weighted, or re-derived any figures from the original publication; Tables 1 and 2 are adapted directly from it, as noted in the table notes.

3. Results and Discussion

3.1 Site-Level Sample Composition

The two study sites, though both classified as slum settlements, differed in composition in ways that seem worth dwelling on rather than passing over. Male respondents made up a slight minority of the Dhaka sample (45.34%) but a clear majority in Tongi (56.18%), with the reverse true for female respondents (Table 1). This is not, on its own, a dramatic difference — but it is not nothing either, and it lines up reasonably well with what one would predict from the two local economies: Tongi's role as a center of ready-made garment manufacturing has historically drawn a distinct gender composition of migrant labor compared with Dhaka's more heterogeneous day-labor and informal-trade economy (Adams et al., 2020). Marital status distributions, by contrast, were strikingly similar across sites — married respondents made up roughly three-quarters of both samples (76.95% in Dhaka, 78.41% in Tongi), with widowed and separated respondents forming small but comparable minorities at each site (Table 1). If anything, this convergence is the more interesting finding: it suggests that whatever forces shape care-seeking behavior in these settlements, marital structure is probably not among the strongest site-differentiating factors, even where labor-market composition clearly is.

3.2 Poverty Probability and Its Distribution

Table 1: Sex and Marital Status Distribution of Working-Age Slum Dwellers in Dhaka and Tongi, Bangladesh (2013–2014). This table reports the weighted percentage distribution of respondent sex and marital status (unmarried, married, widowed, separated) among adults aged 15–64 years who reported a chronic illness lasting three months or longer, stratified by study site. The Dhaka sample comprised 509 respondents and the Tongi sample 536 respondents. Figures are adapted, without alteration, from the original cross-sectional survey reported by Adams et al. (2020). (N = 509 for Dhaka; N = 536 for Tongi. Adapted from Adams et al. (2020)).

Table 2: Poverty Probability Index (PPI) Mean Score and Quintile Distribution Among Working-Age Slum Dwellers in Dhaka and Tongi, Bangladesh. This table presents the mean PPI score — a standardized proxy for the likelihood that a household falls below the national poverty line — together with the weighted percentage of respondents in each of the five PPI quintiles, ranked from poorest (1st) to least poor (5th), at each study site. Higher first-quintile shares indicate a larger proportion of the most economically vulnerable households. Figures are adapted, without alteration, from Adams et al. (2020). 

Poverty Probability Index scores told a somewhat different story. Tongi's mean PPI score (57.48%) was noticeably higher than Dhaka's (49.03%) (Table 2), which — if we take PPI as a reasonably faithful proxy for the likelihood of living below a given poverty line — implies a poorer overall population in Tongi despite its identity as an industrial hub. This is, admittedly, a little counterintuitive at first glance; one might expect factory employment to correlate with somewhat more stable, if still modest, household income. It is possible that the explanation lies less in income level than in income volatility or in cost-of-living differences between the two sites, though the available data do not let us adjudicate this cleanly, and we would caution against over-reading a single summary statistic. The quintile breakdown adds some texture: both sites showed their largest concentration of households in the fourth and fifth quintiles — the more economically secure end of the distribution — but Tongi's first-quintile share (19.36%) exceeded Dhaka's (13.48%) by a meaningful margin (Table 2), suggesting a somewhat more polarized poverty profile in Tongi, with larger shares at both the most and least vulnerable ends simultaneously.

3.3 Situating These Patterns Within the Broader Literature

Taken together, these descriptive contrasts resonate with several threads from the literature reviewed earlier. The site-level heterogeneity documented here echoes Dodd et al.'s (2019) broader observation that primary care systems across the Asia-Pacific region struggle in part because they are designed around generic, rather than locally differentiated, population profiles — a one-size-fits-all catchment model applied to populations that, as this analysis suggests, may not be especially similar even within the same national health system. It also lends some support to the distinction Baum et al. (2017) draw between comprehensive and selective care models: a poverty-polarized population like Tongi's, with substantial minorities at both extremes of economic vulnerability, seems poorly served by a narrow, single-condition selective model and would more plausibly benefit from the kind of whole-population, coordinated approach that comprehensive primary care aims for.

We would also note, with some caution, a possible connection to the migrant health screening inconsistencies documented by Hjern et al. (2019) in the European context. Tongi's role as an export-processing hub means its population likely includes a meaningful share of internal labor migrants, a group whose health-seeking patterns may not map cleanly onto more settled slum populations elsewhere — though this is, we should be honest, an inference rather than something the underlying dataset was designed to test directly.

3.4 Implications for Chronic Disease Follow-Up

If there is a unifying implication across these observations, it is this: chronic illness care in these settlements cannot be treated as a single undifferentiated problem, even within the narrow category of "urban slum." The site differences documented in Tables 1 and 2 are modest in magnitude but plausible in mechanism, and they suggest that follow-up and self-management interventions — of the kind piloted successfully elsewhere by Lorig et al. (1986, 1999) and later scaled through the NHS Expert Patients Programme (Halligan et al., 2006) — would likely need at least some local tailoring rather than a uniform national rollout. The partnership-based model described by Russell et al. (2019), which explicitly incorporates vulnerable-community members into intervention design rather than designing around them, seems like a reasonable template for that kind of tailoring, though we recognize this is a suggestion drawn by analogy rather than a tested recommendation.

4. Proposed Policy and Information-Systems Framework

Building on the patterns discussed above and the wider literature, the following measures are proposed as directions for future piloting and testing — not as validated interventions:

• Stronger governance around chronic and contagious illness care, including clearer enforcement mechanisms tied to legally defined standards of practice.

• A structured, longitudinal record of patient illness and medication history, so that missed follow-up visits can be identified and addressed rather than simply lost to attrition.

• Pharmacy- and medical-store-level accountability, including a requirement to record dispensing information against a prescribing clinician's reference.

• Community-based reminder systems, potentially delivered through local media or trusted community channels, to re-engage patients who have drifted from consistent care.

Figure 1 translates these four measures into a proposed entity-relationship (ER) database schema, linking patient,

Figure 1: Proposed Entity-Relationship (ER) Diagram for a Longitudinal Chronic-Illness Information System. This diagram depicts the database schema underlying the information-systems framework proposed in Section 4, linking eight entities — Patient, Hospital, Doctor, Medical Record, Prescription, Medical Store, Legal_Issue, and Medical_Test — each shown with its primary key (underlined) and foreign keys. Directional arrows and cardinality labels (e.g., 1..*) indicate how a patient's record connects to prescribing clinicians, dispensed medication, diagnostic tests, and any associated legal or regulatory action. The schema is intended to enable continuous, cross-visit tracking of patient care, pharmacy accountability, and enforcement of care standards. It is a conceptual design proposed for future piloting and has not yet been implemented or empirically validated.

hospital, doctor, prescription, medical-store, legal-issue, and diagnostic-test records within a single longitudinal information system designed to operationalize the governance, follow-up, and accountability mechanisms described above.

5. Limitations

Several limitations bear directly on how this paper's contribution should be read. First, and most fundamentally, this is a secondary analysis: we did not collect, weight, or independently verify the underlying survey data, and any error or bias present in Adams et al. (2020) — including the recall bias inherent in self-reported chronic illness, which the original authors themselves acknowledge — necessarily carries through to the interpretations offered here. Second, because household-level sampling permitted multiple eligible respondents per household, observations within the original dataset are not fully independent, which limits how confidently one can generalize the descriptive contrasts in Section 3 beyond simple description. Third, the underlying fieldwork was conducted in 2013–2014; more than a decade of urban change in both Dhaka and Tongi means the site-level patterns discussed here should be treated as historically informative rather than necessarily current. Finally, the policy framework proposed in Section 4 is derived from literature synthesis and descriptive pattern-matching rather than from piloted intervention data, and should be understood as a proposal for future testing rather than a validated solution.

6. Conclusion

This paper set out, deliberately, as a secondary analysis rather than an original data collection — and that framing matters for how its contribution should be weighed. Revisiting Adams et al.'s (2020) descriptive findings from Dhaka and Tongi alongside the wider primary care and chronic disease self-management literature reveals modest but plausible site-level differences in gender composition and poverty profile, differences that seem to track underlying economic structure more than any single demographic factor. These patterns argue, we think reasonably, against a uniform national approach to slum chronic-illness care and toward more locally tailored, partnership-based models of the kind piloted elsewhere. What is missing, more than anything, is longitudinal infrastructure: a way of following patients across visits rather than capturing them once. Strengthening that infrastructure, alongside continued primary care governance reform, seems a more realistic near-term goal than any single intervention, however well designed.

References