Spatio-temporal analysis of Plasmodium falciparum prevalence to understand the past and chart the future of malaria control in Kenya

Abstract

Background: Spatial and temporal malaria risk maps are essential tools to monitor the impact of control, evaluate priority areas to reorient intervention approaches and investments in malaria endemic countries. Here, the analysis of 36 years data on Plasmodium falciparum prevalence is used to understand the past and chart a future for malaria control in Kenya by confidently highlighting areas within important policy relevant thresholds to allow either the revision of malaria strategies to those that support pre-elimination or those that require additional control efforts.

Methods: Plasmodium falciparum parasite prevalence (PfPR) surveys undertaken in Kenya between 1980 and 2015 were assembled. A spatio-temporal geostatistical model was fitted to predict annual malaria risk for children aged 2-10 years (PfPR_2-10) at 1 × 1 km spatial resolution from 1990 to 2015. Changing PfPR_2-10 was compared against plausible explanatory variables. The fitted model was used to categorize areas with varying degrees of prediction probability for two important policy thresholds PfPR_2-10 < 1% (non-exceedance probability) or ≥ 30% (exceedance probability).

Results: 5020 surveys at 3701 communities were assembled. Nationally, there was an 88% reduction in the mean modelled PfPR_2-10 from 21.2% (ICR: 13.8-32.1%) in 1990 to 2.6% (ICR: 1.8-3.9%) in 2015. The most significant decline began in 2003. Declining prevalence was not equal across the country and did not directly coincide with scaled vector control coverage or changing therapeutics. Over the period 2013-2015, of Kenya's 47 counties, 23 had an average PfPR_2-10 of < 1%; four counties remained ≥ 30%. Using a metric of 80% probability, 8.5% of Kenya's 2015 population live in areas with PfPR_2-10 ≥ 30%; while 61% live in areas where PfPR_2-10 is < 1%.

Conclusions: Kenya has made substantial progress in reducing the prevalence of malaria over the last 26 years. Areas today confidently and consistently with < 1% prevalence require a revised approach to control and a possible consideration of strategies that support pre-elimination. Conversely, there remains several intractable areas where current levels and approaches to control might be inadequate. The modelling approaches presented here allow the Ministry of Health opportunities to consider data-driven model certainty in defining their future spatial targeting of resources.

Keywords: Kenya; Malaria; Model-based geostatistics; Plasmodium falciparum.

Fig. 1

Kenya’s counties and populated malaria risk margins: 47 counties shown as dark lines with the extents of major rivers and lakes (light blue); areas unable to support Plasmodium falciparum transmission (dark grey) and low population density (light grey). Turkana (1), West Pokot (2), Trans Nzoia (3), Bungoma (4), Busia (5), Kakamega (6), Siaya (7), Kisumu (8), Homa Bay (9), Migori (10), Kisii (11), Narok (12), Bomet (13), Nyamira (14), Kericho (15), Vihiga (16), Nandi (17), Uasin Gishu (18), Elgeyo Marakwet (19), Baringo (20), Nakuru (21), Nyandarua (22), Laikipia (23), Nyeri (24), Murang’a (25), Kiambu (26), Nairobi (27), Kajiado (28), Makueni (29), Machakos (30), Embu (31), Kirinyaga (32), Tharaka Nithi (33), Meru (34), Samburu (35), Isiolo (36), Marsabit (37), Mandera (38), Wajir (39), Garissa (40), Lamu (41), Tana River (42), Kitui (43), Taita Taveta (44), Kwale (45), Kilifi (46), Mombasa (47). To establish the likely margins of malaria transmission, a temperature suitability index (TSI) has been used based on the monthly average land surface temperatures, the average survival of Anopheles mosquitoes and the length of sporogony that must be completed within the lifetime of one Anopheline generation, where 0 represents the inability to support transmission (dark grey) [14]. Kenya’s population is unevenly distributed within its national borders, with large areas of its land mass characterized by unpopulated areas represented by large conservation areas and deserts. Areas where population density is less than 1 person per km² (light grey) [11] (Fig. 1)  were excluded from subsequent malaria risk extraction

Fig. 2

Annual predicted posterior mean community Plasmodium falciparum parasite rate standardized to the age group 2–10 years (PfPR_2–10) at 1 × 1 km spatial resolution from 1990 to 2015 ranging from zero (dark blue) to 93% in 2003 (dark red) in Kenya. The corresponding standard errors are provided in the Additional file 5

Fig. 3

The national annual mean (black line), 2.5–97.5% (light green boundaries) interquartile credibility range (ICR) and 25–75% ICR (dark green boundaries) of the posterior PfPR_2–10 predictions in Kenya from 1990 to 2015. Unsuitable areas for malaria transmission and those with very low population were excluded in the computation of mean PfPR_2–10 and ICR. Major malaria timelines are shown in bottom panel. Blue boxes represent changing first line anti-malarial treatment and diagnostic policies using malaria rapid diagnostic tests (mRDT). Green boxes represent changing approaches to the delivery of insecticide-treated nets (ITN) through to the provision of free-of-charge of long-lasting insecticide-treated nets (LLIN) during mass campaigns in 2006, 2008, 2011/12, 2014 and 2015 alongside sustained routine delivery to infants and pregnant mothers at clinics. Indoor Residual Spraying (IRS), ( yellow boxes), has been targeted to different counties since 2006 starting in focal areas of 12 counties, by 2010/11 expanding to 16 epidemic prone and 4 endemic counties, and stopped in 2013. Peach colored boxes represent periods of drought while red represents excessive El Niño rainfall, all classified as national disasters

Fig. 4

Annual county level average mean PfPR_2–10 values in populated areas 2013–2015 classified as < 1%, 1–4%, 5–9%, 10–29%, ≥ 30%

Fig. 5

Composite of 3 years 2013, 2014 and 2015 showing areas where predicted PfPR_2–10 is less (non-exceedance probability) than 1% which were > 80% confidently predicted (light green and dark green) or > 90% confidently predicted (dark green); and areas where PfPR_2–10 is greater (exceedance probability) than 30% which were > 80% confidently predicted (light red and dark red) or > 90% confidently predicted (dark red). Areas which do not support malaria transmission are shown in grey (see Fig. 1); all other areas where transmission can occur is shown in white