Masood Ali Shaikh  ( Independent Consultant, Karachi, Pakistan. )

Objective: To analyse the geospatial and temporal distribution of disease and injury burden and factors that influence health care seeking behaviour in thirty-six districts of Punjab in Pakistan.

 

Methods: Utilizing the district level-Pakistan Social and Living Standards Measurement survey data for 2010-2015 from the Pakistan Bureau of Statistics and Human Development Index (HDI) data for the years 2011-2015 from the United Nations Development Programme; choropleth maps were created based on the district level proportions of population reporting having been sick or injured and having sought health consultations for all three waves of PSLM surveys and based on HDI for the years 2011, 2013 and 2015 for the Punjab districts. Spatial cluster analysis of having been sick or injured and having sought health consultations for three waves of PSLM was conducted.

 

Results: For the years (2010-11), (2012-13) and (2014-15) respectively; the mean proportions of population that reported being sick or injured were (7.36±2.23), (6.99±2.33), and (5.70±1.72). The corresponding mean proportions that sought health consultation were (96.11±2.92), (96.05±2.26) and (96.82±2.15) respectively. Having sought health consultation in years (2010-11) and (2012-13), having been sick/injured in (2010-11), and human development index of 2013 as well as 2015 were statistically significant determinants of having sought consultation in the last wave of (2014-15) (p<0.05).

 

Conclusion: Findings show a decline in the reported disease and injury burden between 2010 and 2015 but constant rates of seeking health care. HDI and having sought care previously are the major determinants of subsequent health consultation in Punjab. Future studies need to focus on how these results can be utilized to health inequalities in Pakistan.

 

Keywords: Health, Injury, Pakistan, Referral and Consultation, Geographic Information Systems. (JPMA 72: 1913; 2022)

 

DOI: https://doi.org/10.47391/JPMA.22-99

 

 

Introduction

 

Spatial distribution of disease and health seeking behaviour is important for understanding how place and health are intertwined.^1,2 Disease mapping, clustering, and trends over place and time for burden of disease and health seeking behaviour can help in crafting evidence-based public policies for improving population health.³  Pakistan was classified as lower middle-income country by the World Bank in 2018.⁴  Public sector health facilities tend to be the primary source of curative healthcare services for large proportion of populations in many low- and middle-income countries.⁵  Pakistan's Federal Ministry of Health was dissolved in 2011, with responsibility for health services planning and provision being given to the country's provinces.⁶  With provincial governments being primarily responsible for health in the public sector, it is imperative to understand the spatial distribution of health and health-seeking behaviour at the provincial level.

Punjab province has 36 districts with city of Lahore — also a district — as its capital. Previous studies in Pakistan have examined the spatial distribution of the general disease and injury burden in the country.^7,8 However, since devolution of provinces, they have been determining health policies and allocating budgets for healthcare services including operating and the managing of public sector health facilities within their geographic boundaries. It is imperative to study spatio-temporal distribution, clustering of overall disease burden and health seeking proportions, and their determinants at provincial level in Pakistan.

Provinces, districts, tehsils, and union councils are respectively, the first, second, third, and fourth level of administrative subdivisions in the country. The Pakistan Social and Living Standards Measurement (PSLM) surveys are conducted by the Pakistan Bureau of Statistics (PBS) that provide representative district level indices.⁹  As of November, 2020, three waves of surveys (2010-11, 2012-13, and 2014-15) have been completed, for which district level data have been freely available for download at the PBS official website. The United Nations Development Programme (UNDP), Pakistan, released 'Pakistan Human Development Index Report 2017' that provides district level 'Human Development Index' (HDI).¹⁰

The Pakistan district level HDI is a composite indicator based on two health indicators i.e. immunization rate (percentage of the children aged 12 to 23 months who have been fully immunized), and the satisfaction with health facility (if any of household members had responded as having lack of access to quality healthcare facility, owing to either its availability, cost, suitability, or ill-equipped status); both were taken from PSLM. Additional indicators used in the calculation of this composite measure were mean years of schooling, expected years of schooling, and living standards from the multidimensional poverty index comprising of electricity, drinking water, sanitation, infrastructure, household fuel use, and household assets; taken from PSLM and other surveys conducted in Pakistan.

Punjab is the largest of the four provinces in Pakistan, in terms of population. With a total population of 73,621,290, based on the last census conducted in year 2017; representing 52.95% as a proportion of total population in the country. The objectives of study were three-fold: (1) To analyze spatial and temporal distribution of having been sick or injured, and having sought health consultation for it in the three waves of PSLM for Punjab; (2) To determine the spatial clustering of overall disease and injury burden and health seeking proportions in the three waves of PSLM; and (3) To establish the determinants of having sought health consultation in the 2014-15 wave of PSLM; using indices of having been sick/injured in the all three waves of PSLM, having sought health consultations in the first and second waves of PSLM, and the 2011, 2013 and 2015 district level Human Development Index (HDI) in Punjab.

The hypothesis for this secondary analysis was to study the relationships between having sought health consultations in 2014-15 and positively impacted by having been sick in 2014-15. In addition, having been sick or injured and having sought health consultations in the years 2010-11, and 2012-13. It was hypothesised that a high district level HDI would result in higher proportions of seeking health consultations in the year 2014-15.

 

Methodology

 

The PSLM survey data for all waves are freely available as PDF files on the Pakistan Bureau of Statistics website. PBS has rigorous data quality control checks in place, that are also explained on their website in detail. The PDF files of district level PSLM data were downloaded for the three rounds i.e. 2010-11, 2012-13, and 2014-15, from the Pakistan Bureau of Statistics website. The Punjab province data for the three survey waves were entered into a spreadsheet programme Excel, pertaining to district level proportion/percentage of all-age population that either fell sick or injured during the last weeks prior to the conduct of interviews, and on health consultation sought for them. The spreadsheet was joined with Punjab districts shapefile in ArcGIS 10.7. Similarly, the district level HDI data for the years 2011, 2013 and 2015 were downloaded from the UNDP website, entered into a spreadsheet, and joined with Punjab districts shapefile.

Choropleth maps i.e. maps in which areas are colour-coded based on the variable that is being mapped, were created based on the district level proportions of population reporting having been sick or injured and having sought health consultations for all three waves of PSLM surveys. Similarly, choropleth maps were created based on HDI for the years 2011, 2013 and 2015 for the Punjab districts.

Spatial cluster analysis of having been sick or injured and having sought health consultations for three waves of PSLM was conducted. Global Moran's I values were calculated, which tests for the presence of spatial association or autocorrelation in the entire study area; with 999 permutations for the calculation of pseudo p-values. Followed by Local Indicators of Spatial Association (LISA) maps, which represent the presence and type of spatial clustering and their statistical significance at the sub-divisions studied e.g. districts. Finally, Ordinary Least Squares regression and Spatial Lag regression models were developed with having sought health consultations in the third wave of PSLM i.e. 2014-2015, as dependent variable. While having sought health consultations in the PSLM surveys of 2010-11 and 2012-13, having been sick or injured in all three waves of PSLM, and HDI for the years 2011, 2013 and 2015, as explanatory variables. All mapping and statistical analysis was done using GeoDA 1.14, an Open-Source GIS software.

 

Results

 

In the first wave of PSLM i.e. 2010-11, the proportion of population reported being sick or injured by district ranged from 3.85 to 12.24 (mean: 7.36±2.23, in 2012-13 wave of PSLM this proportion ranged from 2.63 to 12.46 (mean: 6.99±2.33), and in the 2014-15 wave of PSLM the proportion ranged from 2.64 to 9.71 (mean: 5.70±1.72). The 36 districts of Punjab and their names are shown in the map as Figure-1.

The spatial distribution of population proportion reported having been sick or injured in three waves of PSLM are shown in Figure-2. In the 2010-11 PSLM survey, the proportion of population reported having sought health consultation for their sickness/injury ranged from 86.20 to 99.47 (mean: 96.11±2.92, in 2012-13 survey this proportion ranged from 89.16 to 99.79 (mean: 96.05±2.26), while in the last i.e. 2014-15 survey it ranged from 90.76 to 100 (mean: 96.82±2.15). The spatial distribution of population proportion reported having been sought health consultation for their sickness or injury in three waves of PSLM are shown in Figure-3.

The map legends of Figures-2 and 3 show the categories grouped together with similar breaks in all three maps for comparisons across the survey waves. The spatial distribution of Human Development Index (HDI) score by district for the years 2011, 2013, and 2015 are shown in figure 4. In 2011, the HDI ranged from 0.399 to 0.824 (mean: 0.614±0.106), in 2013 this score ranged from 0.481 to 0.858 (mean: 0.690±0.085), while in the year 2015 it ranged from 0.506 to 0.877 (mean: 0.710±0.083). All three maps show the HDI scores grouped in similar categories Figure-4.

Table-1 shows the results of Moran's I test for the three waves of PSLM, for population proportion reported having been sick or injured, and having sought health consultation for it. The results of Local Indicators of Spatial Association (LISA), in terms of cluster and their significance maps of population proportion reported having been sick or injured, for the three waves of PSLM are shown as Figure-5. While the results of Local Indicators of Spatial Association (LISA), in terms of cluster and their significance in maps of population proportion reported having sought health consultations for their sickness or injury, for the three waves of PSLM are shown as Figure-6.

Ordinary least squares (OLS) regression analysis was done next, with the outcome variable as having sought health consultation for sickness or injuries in the third (2014-15) wave of PSLM, with seven explanatory variables i.e. having been sick or injured reported in PSLM 2014-2015, having been sick or injured reported in PSLM 2010-2011, having been sick or injured reported in PSLM 2013-2014, having sought health consultation for sickness/injury reported in PSLM 2010-2011, having sought health consultation for sickness/injury reported in PSLM 2012-2013, district human development index 2011, 2013 and 2015. Owing to high multicollinearity condition number of 208.146, in the explanatory variables in OLS model, spatial lag regression was performed next, using the same outcome and explanatory variables. Regression diagnostics were checked and found to be satisfactory in terms of appropriateness of the spatial lag model. The R-square value for OLS regression was 58.5%, while for spatial lag model it was 64.9%. Hence spatial lag model explained 6 percent more variation in the outcome Figure-2.

Table-2 shows the coefficients and their probability values for OLS and spatial lag regression models. The OLS regression model had five explanatory variables statistically significant at the less than five percent level i.e. having sought health consultation in 2010-11, having sought health consultation in 2012-13, having been sick/injured in 2010-11, and human development index of 2013 as well as 2015. While the spatial lag regression model had the same explanatory variables statistically significant. The spatial lag coefficient in the spatial lag model was also statistically significant at the less than five percent level. The Moran's I of the residuals was calculated for both models. The OLS regression model Moran's I test statistic was -0.169 with pseudo p-value after 999 permutations was 0.068. While the Moran's I test statistic was -0.120 for the spatial lag model with pseudo p-value of 0.201.

 

Discussion

 

This is the first study in Pakistan that assessed the spatial and temporal distribution of having been sick or injured and having sought health consultation for it in the three waves of PSLM for the Punjab province. The spatial distribution of having been sick or injured in the three waves of PSLM surveys clearly show that from 2010-11 to 2014-15 proportion of district populations reporting general disease/injury burden has been progressively decreasing. Whereas this trend might be a true reflection of declining disease and injury burden, on the other hand it may suggest increasing use of self-care/medication. In 2010-11, there were 11 districts that reported sick/injury proportions above 9%, and 7 districts that reported such proportions at or below 5%. In 2012-13, 6 districts reported such proportions above 9%, and 9 districts reported at or below 5%. While in 2014-15, only 2 districts reported such proportions above 9%, and 15 districts reported at or below 5%. The Moran's I value ranged from 0.179 to 0.453, which is weak autocorrelation, but it was statistically significant for all three surveys.

The LISA cluster maps show that in 2010-11 survey there were 6 districts with High-High clusters in the western and central part of the province, and 2 districts with Low-Low clusters in the northern part of the province; in 2012-13 survey there were 3 districts with High-High clusters in the south-western part, and 6 districts with Low-Low clusters in the northern part; while in 2014-15 there were 3 districts with Low-clusters in the north and eastern part, one district with Low-High cluster in the south-central, and one district with High-High cluster in the central part of the province.

The spatial distribution of having sought health consultation for sickness or injured in the three waves of PSLM surveys clearly show that from 2010-11 to 2014-15 the mean proportion of district populations reporting general disease/injury burden is high but has generally remained stagnant between 2010-2015.

In 2010-11, as well as 2012-13, there were three districts that reported having sought health consultations at below 92.5% level, while in 2014-15 there were two such districts. The Moran's I value ranged from -0.102 to 0.202, which is very weak autocorrelation, but it was statistically significant for only 2010-11 survey. The LISA cluster maps show that in 2010-11 survey there was 1 district with High-High, two districts with Low-Low, and one district each with High-Low and Low-High clusters, in the north, central and eastern parts of the province. In 2012-13 survey, there was one district each with High-High and Low-Low clusters, and two districts with Low-High clusters, in the north and western parts, while in 2014-15 there was one district each with Low-High and High-Low clusters in the eastern part of the province. These findings imply that some diseases and injuries still go unregistered in the formal health sector. Similar to other settings where access to health facilities have limited accessibility,¹¹  the use of traditional medicine, alternative therapies and self-medication with over-the-counter drugs are contributing factors to poor health-seeking behaviour in Pakistan.¹²

UNDP Pakistan classified HDI as 'Low' in the range of 0.200 to 0.499; five districts were classified as low in 2011, one district in 2013, and none in 2015. While 0.700 to 0.799 range is classified as 'Medium', and 0.800 to 0.899 range was classified as 'High Medium'. In 2011, eight districts fell in the medium and high-medium groups, 14 in 2013, and 22 in 2015. HDI between 0.500 to 0.699 was classified as 'Low Medium'. While below 0.200 was classified as 'Very Low' and above 0.899 as 'High'. None of the districts fell in low or high categories in any of the three years.

Ordinary least squares (OLS) regression was used to determine the association of having sought health consultation reported in PSLM 2014-15 with eight explanatory variables of having sought health consultation in 2010-11 and 2012-13, having been sick/injured in 2010-11, 2011-13, and 2014-15, and HDI scores in 2011, 2013, and 2015. With the exception three explanatory variables of having been sick/injured in 2012-13, 2014-15, and 2011 HDI, all other five variables were statistically significant at the level of <0.05. However, owing to substantially high multicollinearity in the OLS model, a Spatial Lag (SL) regression model was performed that successfully accounted for it. As spatial/geographical data tends to exhibit spatial autocorrelation, with resultant false correlations; one method for accounting this autocorrelation is the use of spatial lag model. In SL model, the lag variable was statistically highly significant, implying that higher coefficient values in the OLS model partly attributable to multicollinearity. With the exception of having sought health consultation in 2010-11, same explanatory variables were statistically significant in spatial lag regression model as well, albeit with much lower p-values. The variance (R-Squared) explained by the explanatory variables in the SL model was 64.9%, which is 6.4% points higher compared to OLS model.

Regarding associations between statistically significant explanatory variables with the outcome variable in the SL model; on average, each percent point increase in having sought health consultations in 2012-13 was associated with an increase of 0.374 percent points in having sought health consultations in 2014-15, while the effects of other explanatory variables constant; each percent point increase in having been sick or injured in 2010-11 was associated with a decrease of 0.622 percent points in having sought health consultations in 2014-15. In addition, each unit increase in 2013 HDI was associated with a decrease of 0.343 percent points in having sought health consultations in 2014-15; and finally, each unit increase in 2015 HDI was associated with an increase of 0.319 percent points in having sought health consultations in 2014-15. Other studies have documented wealth index at individual and community levels as principal determinant of health-seeking behaviour in Pakistan.¹³  Interestingly, having been sick/injured in 2014-15 did not show statistically significant associations with having sought health consultations in 2014-15.

The lag coefficient was negative in the SL model and statistically significant.

Regarding limitations of this study, the lowest level at which PSLM data are available is the district, that would ostensibly mask the sub-district i.e. tehsil and union council level proportions and trends.

Data on accessibility, affordability, and acceptability of health facilities, income level, and educational attainment are not available in the PSLM surveys. However, some of the HDI indicators are based on PSLM attributes listed previously that potentially serve as proxy for some of these attributes. Finally, this secondary analysis was limited to total i.e. all age population, cumulative for both genders, and for urban rural populations.

A more nuanced analysis based on children, male/female, and urban/rural disaggregation might provide deeper insights for better informed health care policy crafting. Future studies using PSLM data need to compare this study's cumulative analysis with more nuanced analysis to decipher spatial distribution, clustering and healthcare consultation trends and their correlates in Punjab province.

 

Conclusion

 

The findings show a decline in the reported disease and injury burden between 2010 and 2015 but constant rates of seeking health care. Human development index and having sought care previously were the major determinants of subsequent health consultation in the studied settings. Taken together the results provide useful information for health planners in estimating population health needs over large areas. The data presented here would be useful when accompanied with nuanced individual and contextual data which ascertains actual health seeking choices and the reasons behind them. Future studies need to evaluate how the determinants of health service utilization can be applied to address health care inequalities among vulnerable populations in the Punjab province of Pakistan.

 

Disclaimer: None.

 

Conflict of Interest: None.

 

Funding Disclosure: None.

 

References

 

1.      Proud Tembo NF, Bwalya Muma J, Hang'ombe B, Munyeme M. Clustering and spatial heterogeneity of bovine tuberculosis at the livestock/wildlife interface areas in Namwala District of Zambia. Vet World. 2020; 13:478-88.

2.      Shaikh MA, Malik NA. Spatial Cluster Analysis of New And Relapsed Cases Of Pulmonary Tuberculosis By District: Pakistan 2015. J Ayub Med Coll Abbottabad. 2019; 31:293-5.

3.      Wahidin M, Hidayat MS, Arasy RA, Amir V, Kusuma D. Geographic distribution, socio-economic disparity and policy determinants of smoke-free policy adoption in Indonesia. Int J Tuberc Lung Dis. 2020; 24:383-9.

4.      World Bank Country and Lending Groups. [Online] [Cited 2019 February 21]. Available from: URL: https://datahelpdesk.worldbank.org/knowledgebase/articles/90 6519-world-bank-country-and-lending-groups.

5.      Asamani JA, Chebere MM, Barton PM, D'Almeida SA, Odame EA, Oppong R. Forecast of Healthcare Facilities and Health Workforce Requirements for the Public Sector in Ghana, 2016-2026. Int J Health Policy Manag. 2018; 7:1040-52.

6.      Zaidi SA, Bigdeli M, Langlois EV, Riaz A, W Orr D, Nasir Idrees N, et al. Health systems changes after decentralisation: progress, challenges and dynamics in Pakistan. BMJ Glob Health. 2019; 4:e001013.

7.      Shaikh MA. Correlates of being sick or injured in under five-year old children by district in Pakistan: A spatial analysis case study. J Pak Med Assoc. 2018; 68:633-8.

8.      Shaikh MA. Spatial distribution of being sick or injured in the past two weeks by district in Pakistan. J Pak Med Assoc. 2016; 66:1648- 51.

9.      Pakistan Bureau of Statistics. Pakistan Social and Living Standards Measurement. [Online] 2020 [Cited 2020 February 16]. Available from: URL: http://www.pbs.gov.pk/content/pakistan-social-andliving- standards-measurement.

10.    UNDP Pakistan. Pakistan Human Development Index Report 2017. [Online] 2017 [Cited 2019 February 13]. Available from: URL:https://www.undp.org/content/dam/pakistan/docs/HDR/HD I%20Report_2017.pdf.

11.    Pathak A, Agrawal N, Mehra L, Mathur A, Diwan V. First Aid Practices and Health-Seeking Behaviors of Caregivers for Unintentional Childhood Injuries in Ujjain, India: A Community- Based Cross-Sectional Study. Children. 2018; 5:124-9.

12.    Zahra T, Rashid A, Ilyas A, Sohail A, Baig S. Curative Health Seeking Behavior in Pakistan. Prof Med J. 2019; 26:41-7.