Accuracy of recording of malaria rapid diagnostic test results in Uganda

Abstract

Background

Malaria rapid diagnostic tests (RDTs) have been critical in promoting the rational use of antimalarials and strengthening malaria surveillance. However, the accuracy of routinely reported RDT results in Uganda remains unclear. The study’s objective was to measure the level of agreement between healthcare workers (HCWs) and an external panel’s RDT results among lower-level public health facilities in Busoga and Lango regions, Uganda.

Methods

A prospective study was conducted in 16 public health facilities in four purposively selected districts in Uganda. At each study site, images of all RDTs were taken as soon as the HCW had finished interpreting the test results and uploaded into HealthPulse (Audere, Seattle, WA USA), a digital RDT reader. Corresponding patient data was captured from the outpatient department (OPD) register, including demographics and RDT test results. RDT images were sent to a trained, external panel for interpretation. Cohen’s kappa statistic (κ) was used to determine agreement. The associations between characteristics of health facilities, HCWs and RDTs and the level of agreement were analysed using meta-analytical approaches.

Results

From June to November 2023, 40,049 RDT images were captured, of which 37,137 (92.7%) were included in the analysis. Overall, the test positivity rate based on OPD records was 61.8%. The overall agreement was strong (κ 0.81, 95% confidence interval [CI] 0.76, 0.85). Where disagreement occurred, HCWs misrecorded more RDT results as positive (7.2%) than negative (1.9%). Agreement was higher in Busoga (κ 0.86, 95% CI 0.83, 0.88) compared to Lango (κ 0.78, 95% CI 0.75, 0.81). Lower agreement levels were also associated with certain RDT products and RDTs with faint lines.

Conclusion

The study found a strong level of agreement between HCWs' RDT results and an external panel. However, significant proportions of results were misrecorded as positive or negative, particularly in the Lango region. Targeted interventions, such as RDT validation exercises and tailored refresher training, are recommended to enhance RDT reporting accuracy in Uganda.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12936-025-05637-7.

Background

Over the past two decades, Uganda has made remarkable progress in malaria control, achieving a 47% decline in malaria incidence and an 81% reduction in mortality between 2000 and 2023 [1]. One of the major contributors to this success has been improvement in malaria case detection and management, largely driven by the promotion and widespread use of malaria rapid diagnostic tests (RDTs) [2]. Despite this progress, Uganda remains a significant contributor to the global malaria burden, with recent World Health Organization (WHO) estimates ranking the country third in malaria cases and tenth in malaria-related deaths in 2023 [1].

RDTs are immunochromatographic test strips that detect parasite antigens or enzymes in blood that are either genus- or species-specific [3]. Since 2011, Uganda adopted the WHO policy of mandatory parasitological confirmation of malaria with either microscopy or RDTs before the prescription of antimalarials [4]. At the inception of the policy, microscopy was to remain the “gold standard” for malaria diagnosis at health centre (HC) III (i.e. health units with laboratory services) and higher-level facilities, while RDTs were limited for use at HC IIs and other facilities whenever microscopy was not possible [5]. However, microscopy is labour-intensive, and its accuracy is dependent on the expertise of the laboratorian, the equipment, and the quality of staining reagents [6]. Given the high patient load amidst staffing shortages, easier-to-use RDTs have become the preferred test type across all facility types, especially at lower-level health facilities (HCs II and III). Increased use of RDTs has boosted testing rates, resulting in a significant decline in the presumptive treatment of malaria among febrile patients [7]. The scale-up of RDTs has also improved the accuracy of reported malaria surveillance data, thereby enabling Uganda to better monitor malaria trends and the effectiveness of interventions [8].

For malaria RDTs to serve their intended purpose, test results must be accurate, healthcare workers (HCWs) must prescribe antimalarials based on these results, and the results must be accurately recorded in health facility registers. Studies investigating compliance of HCWs with malaria RDT results have reported varying levels of adherence [9–13]. A meta-analysis of 14 studies reported an overall compliance with RDT results of 83%, with factors such as patient expectations, HCW cadre, work experience, and perceived test accuracy influencing adherence [14]. Additionally, falsification of RDT results to justify treatment in the absence of an RDT, or in the setting of a negative RDT or human error, may compromise the quality of reported data from health facilities. Mistakes in data reporting are attributed to a conundrum of reasons, including non-commitment to data quality, a laborious documentation process, inadequate supervision, large patient volumes, and insufficient training on data reporting [15, 16].

In Uganda, previous studies have documented high proficiency in administering and interpreting malaria RDTs, including among trained community health workers (CHWs) [17, 18]. However, none of these studies have assessed the accuracy of malaria RDT data recorded in public health facilities. This study aimed to assess the accuracy of RDT results recorded by HCWs by measuring the level of agreement between RDT results recorded in health facility registers and those of a trained external panel at lower-level public health facilities in Uganda’s Busoga and Lango regions.

Methods

Study design

This was a prospective study conducted as part of a larger multi-country study in sub-Saharan Africa, including Benin, Côte d’Ivoire and Nigeria. Given the differing contexts for each of the four countries involved in this study, it was deemed meaningful for each country to write an independent paper expounding on country-specific drivers of agreement. This paper focuses specifically on the Uganda component. A detailed description of the overall study methods and a summary of key results were published in the Malaria Journal under the collection “Diagnosing the Data: Malaria RDT Recording and Reporting Accuracy in Four African Countries and Implications for Surveillance” [19]. In brief, between June and November 2023, high-quality images of malaria RDTs performed at health facilities among patients with suspected malaria cases were captured using an AI-powered digital RDT reader smartphone application (HealthPulse, Audere, Seattle, WA, USA). The images were sent to an external panel for independent interpretation, and the results recorded by HCWs in facility registers were compared to those interpretations to determine the level of agreement. It is important to note, however, that unlike the multi-country overview paper [19], which used multinomial logistic regression to examine factors associated with different types of inaccuracies (RDTs misrecorded as positive versus misrecorded as negative), this Uganda-specific analysis focuses on overall agreement using Cohen’s kappa.

Study sites

The study was conducted in 16 public health facilities (HC II and III) located in two high-burden regions: Busoga and Lango in Eastern and Northern Uganda, respectively. Both are high transmission settings, and based on the 2018 Malaria Indicator Survey, Busoga had the third highest parasite prevalence and Lango the fourth highest in the country [20]. Based on recent estimates by the Malaria Atlas Project, the parasite prevalence is projected to be 26% in Busoga and 32% in Lango [21]. HC II facilities are defined as those at the parish level that provide basic outpatient services and treat common illnesses such as uncomplicated malaria, while HC III facilities offer diagnostic, outpatient, and maternity services at the sub-county level. At HCII facilities, the HCW performs all outpatient duties, including triage, screening, RDT testing, treatment, and recording within the OPD room. In contrast, at HCIII facilities with a laboratory, malaria testing is conducted in the laboratory, and results are sent to the OPD, where the HCW interprets them, prescribes treatment, and records the outcome in the register. In each region, two districts were purposively selected based on two criteria: (1) the presence of HC II or III public facilities, and (2) the absence of ongoing external interventions supporting malaria testing capacity at the facilities. All lower-level public health facilities in the selected districts were considered eligible if they met the following criteria: (1) three years (2020 to 2022) of complete outpatient malaria data (defined as at least 9 of 12 months each year) reported to the District Health Information System 2 (DHIS-2), which serves as Uganda’s health management information system, and (2) a minimum of 50 RDTs performed monthly from 2020 to 2022. Using medians, eligible facilities within each district were grouped into four strata based on patient volume and test positivity rate (TPR). One facility from each stratum was randomly selected, resulting in four study facilities per district (Fig. 1).

Fig. 1.

Map of Uganda showing the study area

Data collection procedures

Trained research assistants (RAs) collected RDT images and data using the HealthPulse application on weekdays. At baseline, a facility survey was conducted to assess infrastructure and resource capacity for malaria case management. Additionally, HCWs were surveyed to collect their demographic information, professional qualifications, and experience using RDTs. RAs observed HCWs as they performed an RDT, using a 19-point standardised checklist to assess proficiency (Supplemental file 1). All HCWs at the study facilities were eligible to participate if they reported current or likely future involvement in performing RDTs or recording their results.

RAs photographed RDTs using the HealthPulse app as soon as practically feasible after the HCW completed interpreting the results, without interfering with or obstructing patient care. The app had an image quality assurance component that immediately flagged images that did not meet quality standards and prompted users to retake the photo when needed. RDT images were matched using barcode labels to corresponding patient data in the OPD registers, which were captured by RAs using the HealthPulse application. HCWs were briefed at the study launch and during the consenting process on the study procedures, including the daily presence of the RAs, their roles and expectations were communicated to ensure minimal disruptions in the routine workflow.

Images were sent to an external panel of trained reviewers who determined whether control and test lines were present in an RDT image. Panellists flagged anomalies in the RDTs and the presence of faint lines. The 3746 images captured during the first four weeks of the study were reviewed by three different panellists, and a near-perfect level of agreement (Fleiss’ kappa = 0.999, p < 0.0001) was measured. Given the high level of interrater agreement, after the first month of the study, each RDT was interpreted by a single, randomly selected member of the panel. The overview paper [19] provides further information on the quality control process for image results.

Data analysis

Characteristics of health facilities, HCWs and patients for whom RDTs were performed were presented as proportions disaggregated by region. The latitude and longitude of health facilities were used to calculate Plasmodium falciparum parasite prevalence for children 2–10 years (PfPR_2-10) within 5 km using the malariaAtlas package in R (R Foundation for Statistical Computing, Vienna, Austria) [22]. HCWs were divided into terciles based on their RDT proficiency scores. Results interpreted by the external panel as positive or invalid but recorded as negative by the HCW were termed ‘misrecorded as negative,’ and those interpreted as negative or invalid by the external panel and positive by the HCW were termed ‘misrecorded as positive.’ The positive predictive value (true positives/[true positives + misrecorded as positive]) and the negative predictive value (true negatives/[true negatives + misrecorded as negative]) were calculated to measure the likelihood that a positive or negative test recorded in a facility register was a true positive or negative, respectively.

Cohen’s kappa was used to measure the level of agreement between malaria RDT results recorded by HCWs in health facility registers and those interpreted by the external panel. Cohen’s kappa provides a single summary measure of agreement adjusted for chance. To generate an overall estimate of agreement across study facilities, a random-effects meta-analytic approach was applied [23]. Specifically, a weighted mean kappa was computed, with each facility-level kappa estimate weighted by the inverse of its variance to account for variation in measurement precision across facilities. In addition to calculating the pooled kappa, potential sources of variability in agreement were explored by including key characteristics of health facilities, HCWs, RDTs and patients on whom RDTs were performed as fixed-effects in the model. All statistical analyses were conducted using R version 4.3.3 [24].

Results

Characteristics of study health facilities

Of the 16 study health facilities, 10 (62.5%) were HCs II, while the remaining six were HCs III (Table 1). The malaria prevalence among children 2–10 years old in the 5 km radius around the health facility was > 30% in 5 (31.3%) health facilities. Most facilities (10, 62.0%) had at least five staff who performed RDTs. Seven facilities (43.8%) had a laboratory technician present. Lango had fewer HCs II (50.0%) compared to Busoga (75.0%). Additionally, more facilities in Lango had a parasite prevalence > 30% (8, 50%) compared to those in Busoga (2, 12.5%).

Table 1.

Characteristics of study health facilities, Uganda 2023

Characteristic	Overall N = 16 n (%)	Region		No. RDTs recorded N = 37,137 n (%)
		Busoga N = 8 n (%)	Lango N = 8 n (%)
District
Otuke	4 (25.0)	0 (0)	4 (50.0)	5830 (15.7)
Dokolo	4 (25.0)	0 (0)	4 (50.0)	17,150 (46.2)
Buyende	4 (25.0)	4 (50.0)	0 (0)	5809 (15.6)
Bugweri	4 (25.0)	4 (50.0)	0 (0)	8348 (22.5)
Health centre level
Level II	10 (62.5)	6 (75.0)	4 (50.0)	20,210 (54.4)
Level III	6 (37.5)	2 (25.0)	4 (50.0)	16,927 (45.6)
Test positivity rate stratum
High	8 (50.0)	4 (50.0)	4 (50.0)	18,604 (50.1)
Low	8 (50.0)	4 (50.0)	4 (50.0)	18,533 (49.9)
RDT volume stratum
High	8 (50.0)	4 (50.0)	4 (50.0)	20,494 (55.2)
Low	8 (50.0)	4 (50.0)	4 (50.0)	16,643 (44.8)
Parasite prevalence (PfPR2-10)
0–9%	1 (6.2)	1 (12.5)	0 (0)	2704 (7.3)
10–19%	1 (6.2)	1 (12.5)	0 (0)	1855 (5.0)
20–29%	9 (56.2)	5 (62.5)	4 (50.0)	14,327 (38.6)
30–39%	5 (31.2)	1 (12.5)	4 (50.0)	18,251 (49.1)
Number of HCWs who perform RDTs
1–2	2 (12.5)	1 (12.5)	1 (12.5)	6836 (18.4)
3–4	4 (25.0)	3 (37.5)	1 (12.5)	7333 (19.7)
5 or more	10 (62.5)	4 (50.0)	6 (75.0)	22,968 (61.8)
Laboratory technician present
Yes	7 (43.8)	3 (37.5)	4 (50.0)	18,464 (49.7)
No	9 (56.2)	5 (62.5)	4 (50.0)	18,673 (50.3)

HCWs: healthcare workers; N: Number; PfPR_2-10: P. falciparum prevalence among children 2–10 years old; RDTs: rapid diagnostic tests

Most RDTs (17,150, 46.2%) were recorded at health facilities in Dokolo district over the study period (Table 1). Health facilities from the high RDT volume stratum performed more RDTs (20,494, 55.2%) than those from the low volume stratum (16,643, 44.8%). Similar numbers of RDTs were performed in facilities with a laboratory technician present (18,464, 49.7%) as without (18,673, 50.3%).

Characteristics of HCWs

Among 217 health facility staff, 171 (78.8%) reported involvement in the administration, recording, and reporting of malaria RDTs and 169 (98.8%) were interviewed. Over the six months of the study implementation, 93 (55.0%) HCWs contributed at least one RDT recording to the analytical dataset. Of these, 46 (49.5%) were female, and a plurality (36, 38.7%) were aged 30–39 years (Table 2). Nurses constituted the largest cadre of HCWs (49, 52.7%), followed by nonmedical or volunteer staff (12, 12.9%) and medical auxiliary staff (10, 10.8%). Most HCWs (76, 81.7%) had a university degree, while more than half (48, 51.6%) had at least 10 years of experience in their professional roles. Most HCWs reported frequently performing RDTs (60, 64.5%).

Table 2.

Characteristics of healthcare workers who recorded rapid diagnostic test (RDT) results during the MaCRA study and number of RDTs recorded, Uganda 2023

Characteristic	Overall N = 93 n (%)	Region		No. RDTs recorded N = 37,137 n (%)
		Busoga N = 49 n (%)	Lango N = 44 n (%)
Sex
Female	46 (49.5)	30 (61.2)	16 (36.4)	16,830 (45.3)
Male	47 (50.5)	19 (38.8)	28 (63.6)	20,307 (54.7)
Age (years)
< 30	22 (23.7)	13 (26.5)	9 (20.5)	5907 (15.9)
30–39	36 (38.7)	18 (36.7)	18 (40.9)	16,098 (43.3)
40–49	26 (28.0)	16 (32.7)	10 (22.7)	9003 (24.2)
50–59	9 (9.7)	2 (4.1)	7 (15.9)	6129 (16.5)
Occupational cadre
Nurse	49 (52.7)	27 (55.1)	22 (50.0)	24,690 (66.5)
Clinical officer	9 (9.7)	4 (8.2)	5 (11.4)	8188 (22.0)
Nonmedical or volunteer staff	12 (12.9)	5 (10.2)	7 (15.9)	2528 (6.8)
Community health worker	8 (8.6)	4 (8.2)	4 (9.1)	97 (0.3)
Lab technician	5 (5.4)	2 (4.1)	3 (6.8)	19 (0.1)
Medical auxiliary staff	10 (10.8)	7 (14.3)	3 (6.8)	1615 (4.3)
Highest educational qualification
Primary or below	3 (3.2)	3 (6.1)	0 (0)	89 (0.2)
Secondary	14 (15.1)	7 (14.3)	7 (15.9)	2271 (6.1)
University	76 (81.7)	39 (79.6)	37 (84.1)	34,777 (93.6)
Years of experience
0–1	8 (8.6)	5 (10.2)	3 (6.8)	2754 (7.4)
2–4	12 (12.9)	8 (16.3)	4 (9.1)	2845 (7.7)
5–9	25 (26.9)	8 (16.3)	17 (38.6)	13,471 (36.3)
10 or more	48 (51.6)	28 (57.1)	20 (45.5)	18,067 (48.6)
Frequency of performing RDTs
Very often (every day)	50 (53.8)	26 (53.1)	24 (54.5)	22,631 (60.9)
Once in a while to often	35 (37.6)	18 (36.7)	17 (38.6)	7908 (21.3)
Never	8 (8.6)	5 (10.2)	3 (6.8)	6598 (17.8)
RDT proficiency score tercile
Low	29 (33.3)	22 (45.8)	7 (17.9)	10,603 (35.4)
Middle	29 (33.3)	15 (31.2)	14 (35.9)	13,054 (43.5)
High	29 (33.3)	11 (22.9)	18 (46.2)	6323 (21.1)

RDTs: rapid diagnostic tests

Compared to Busoga, health facilities in Lango had a higher proportion of male HCWs (63.6% vs. 38.8%) and a greater proportion aged 50–59 years (15.9% vs. 4.1%) (Table 2). In contrast, fewer HCWs in Lango had 10 or more years of experience (45.5% vs. 57.1%). The proportion of HCWs who reported performing RDTs very often was similar between the two regions. However, a larger proportion of HCWs in Lango had RDT proficiency scores in the upper tercile (46.2% vs. 22.9%).

The distribution of RDTs recorded by HCWs was broadly proportional to the frequency of their demographic and professional characteristics. However, several notable deviations were observed. HCWs under 30 years old accounted for 15.9% of RDTs recorded, despite representing 23.7% of the HCWs, whereas those aged 50–59 years recorded 16.5% of RDTs but comprised only 9.7% of HCWs. Nurses and clinical officers recorded a higher-than-expected proportion of RDTs relative to their representation, while other cadres contributed less than expected. HCWs who reported performing RDTs very often made up 64.5% of the total but were responsible for 60.9% of RDTs recorded. Interestingly, HCWs in the middle tercile of RDT proficiency conducted 43.5% of RDTs, compared to 21.1% among those in the highest proficiency tercile.

Characteristics of RDTs observed in this study

During the study period, a total of 45,838 RDT results was recorded in OPD registers. Of these, 40,049 (87.4%) had images captured in the HealthPulse application. Most missing RDT images were performed on weekends or evenings when RAs were not present. After excluding images without corresponding patient data, 37,137 (92.7%) RDT results were included in the final analysis (Fig. 2).

Fig. 2.

Study profile for rapid diagnostic tests observed in the study

The overall TPR among RDTs recorded in the health facility registers was 61.8% (Table 3). The most common RDT product was Bioline Malaria Pf (Abbott, IL, USA), which accounted for 19,152 (51.6%) cases, while STANDARD Q Malaria Pf (SD Biosensor, Gyeonggi-do, Republic of Korea) and Bioline Malaria Pf/Pan each comprised more than 10% of RDTs performed. A total of 4993 (13.4%) RDTs were tagged as having a faint line. Most RDTs were performed on female patients (24,871, 67.0%) and those aged 15 years or more (15,850, 42.7%). There were more patients recorded in OPD registers with a diagnosis of malaria (24,305, 65.4%), of whom 22,914 (61.7%) received antimalarial prescriptions, than the number of positive RDTs recorded (22,952, 61.8%). Close to a third of RDT results (11,242, 30.3%) were recorded in the OPD register by the same individual who performed the RDT.

Table 3.

Characteristics of rapid diagnostic tests observed in the MaCRA study by region, Uganda 2023

Characteristic	Overall N = 37,137 n (%)	Region
		Busoga N = 14,157 n (%)	Lango N = 22,980 n (%)
RDT result recorded in OPD register
Positive	22,952 (61.8)	7722 (54.5)	15,230 (66.3)
Negative	14,185 (38.2)	6435 (45.5)	7750 (33.7)
RDT result interpreted by external panel
Positive	20,913 (56.3)	7124 (50.3)	13,789 (60.0)
Negative	16,107 (43.4)	6976 (49.3)	9131 (39.7)
Invalid	117 (0.3)	57 (0.4)	60 (0.3)
RDT product
Bioline Malaria Pf	19,152 (51.6)	3882 (27.4)	15,270 (66.4)
STANDARD Q Malaria Pf	7289 (19.6)	7036 (49.7)	253 (1.1)
Bioline Malaria Pf/Pana	4304 (11.6)	1 (0)	4303 (18.7)
First Response Malaria Ag (pLDH/HRP2)a	3209 (8.6)	1959 (13.8)	1250 (5.4)
First Response Malaria Pf	2297 (6.2)	520 (3.7)	1776 (7.7)
ParaHIT Malaria Pf	741 (2.0)	741 (5.2)	0 (0)
Bioline Malaria Pf (HRP2/pLDH)a	128 (0.3)	0	128 (0.6)
Other	17 (0)	17 (0.1)	0 (0)
Faint line
Yes	4993 (13.4)	1698 (12.0)	3295 (14.3)
No	32,144 (86.6)	12,459 (88.0)	19,685 (85.7)
Patient sex
Female	24,871 (67.0)	9076 (64.2)	15,795 (68.8)
Male	12,243 (33.0)	5071 (35.8)	7172 (31.2)
Patient age (years)
0–4	8466 (22.8)	4348 (30.7)	4118 (17.9)
5–14	12,819 (34.5)	4291 (30.3)	8528 (37.1)
15 or more	15,850 (42.7)	5518 (39.0)	10,332 (45.0)
Patient diagnosed with malaria
Yes	24,305 (65.4)	7727 (54.6)	16,578 (72.1)
No	12,832 (34.6)	6430 (45.4)	6402 (27.9)
Patient prescribed antimalarial
Yes	22,914 (61.7)	7712 (54.5)	15,202 (66.2)
No	14,223 (38.3)	6445 (45.5)	7778 (33.8)
RDT performed and recorded by the same person
Yes	11,242 (30.3)	6395 (45.2)	4847 (21.1)
No	25,895 (69.7)	7762 (54.8)	18,133 (78.9)

HRP2: histidine-rich protein 2; pLDH: parasite lactate dehydrogenase; RDT: rapid diagnostic test

^aRDT products with two test lines

RDTs from health facilities in Busoga had a lower TPR than Lango (54.5% vs 63.3%) (Table 3). STANDARD Q Malaria Pf was the main RDT product used in Busoga, while Bioline Malaria Pf was the main product in Lango. The age distribution of patients differed by region, with a lower proportion of patients in Lango aged 0–4 years (17.9%) compared with Busoga (30.7%). RDTs in Busoga were more likely to be performed and recorded by the same person (45.2%) compared to Lango (21.1%).

Agreement between the healthcare workers’ and the external panel’s RDT results

Out of the 37,137 RDTs reviewed, 33,743 (90.0%) of the HCW results were in agreement with the external panel results, including 20,254 (54.5%) true positives and 13,489 (36.3%) true negatives. Among the disagreements, 2618 (7.0%) true negatives were misrecorded as positive, 659 (1.8%) true positives were misrecorded as negative, 37 (0.1%) true invalids were misrecorded as negative, and 80 (0.2%) true invalids were misrecorded as positive. The overall proportion of results misrecorded as positive was 7.2% and results misrecorded as negative were 1.9%. The predictive value of a positive RDT result in the health facility register was 88.2% while the predictive value of a negative RDT result in the health facility register was 95.1%. While the TPR declined over the study period in line with seasonal patterns [25], the proportion misrecorded as positive or negative did not vary substantially by study week (Fig. 3).

Fig. 3.

Distribution of malaria rapid diagnostic test results based on external panel interpretation by study week, Uganda, N = 37,020

Cohen's kappa (κ) for the overall level of agreement was 0.81 (95% CI 0.76, 0.85). There was variability in the estimate of κ by health facility, with individual facility-level varying from a low of 0.73 (95% CI 0.69, 0.77) to a high of 0.92 (95% CI 0.90, 0.94) (Fig. 4). Generally, health facilities in Busoga (facilities 1–8) had higher agreement levels compared to those in Lango (facilities 9–16). The level of agreement in Busoga was higher (κ 0.86, 95% CI 0.83, 0.88) compared to Lango (κ 0.78, 95% CI 0.75, 0.81).

Fig. 4.

A forest plot of Cohen's kappa estimates by health facility and the overall estimate from a random-effects model, Uganda 2023

There were a few factors other than region found to be associated with the level of agreement. The two districts of Busoga (Buyende and Bugweri) had higher measures of κ than the two districts of Lango (Otuke and Dokolo) (Fig. 5A). There was no association between either the baseline strata for TPR or RDT volume and the level of agreement. There was a decreasing level of agreement with increasing parasite prevalence, but differences were not statistically significant. Neither the number of staff who performed RDTs nor the presence of a laboratory technician within the facility was associated with the level of agreement.

Fig. 5.

Forest plot of factors moderating Cohen’s kappa score from a mixed-effects model, Uganda 2023. A Health facility characteristics. B Healthcare worker characteristics. C RDT characteristics

Among HCW characteristics, no factor was found to moderate the level of agreement (Fig. 5B). Several RDT characteristics, however, were associated with κ, including RDT product, the number of test lines and the presence of faint lines (Fig. 5C). Compared to First Response Malaria Pf, the level of agreement was significantly lower for First Response Malaria (pLDH/HRP2) (κ 0.73, 95% CI 0.68, 0.79) and higher for STANDARD Q Malaria Pf (0.87, 95% CI 0.83, 0.91). When grouped, the RDTs that include two test lines (Bioline Malaria Pf/Pan, Bioline Malaria Pf (HRP2/pLDH) and First Response Malaria Ag (pLDH/HRP2)) were associated with significantly lower levels of agreement (κ 0.77, 95% CI 0.73, 0.81) than those with a single test line (κ 0.82, 95% CI 0.79, 0.84). Faint test lines were also associated with significantly lower levels of agreement (κ 0.14, 95% CI 0.02, 0.26).

Discussion

In this study, the accuracy of RDT reporting in Ugandan health facilities was evaluated by comparing RDT results recorded in facility registers with results interpreted by an external panel. Overall, there was a strong level of agreement between HCW’s records and the panel’s interpretations, indicating that RDT results reported from health facilities are generally reliable. However, 7.2% of results were misrecorded as positive, resulting in an 88.2% probability that a positive result in the register was a true positive. These findings suggest that while RDT reporting is largely accurate, positive results may be over-reported, potentially obscuring declines in TPR over time and masking the impact of malaria control efforts.

Notable regional discrepancies were observed, with lower agreement levels in the Lango region compared to Busoga. This regional discrepancy may be attributed to several factors. First, most health facilities in Lango experienced a higher patient volume but had relatively fewer HCWs responsible for administering and recording RDT results. A low HCW-to-patient ratio can contribute to reduced accuracy in reporting, as increased workloads and burnout are known to elevate the risk of medical errors [26]. Therefore, efforts to improve data quality and malaria surveillance should prioritise increasing the number of trained HCWs in lower-level health facilities, particularly in high-burden districts to enhance service delivery.

Second, the study also found that RDT products with two test lines, including Bioline Pf/Pan, First Response Pf (pLDH/HRP2) and Bioline Pf (HRP2/pLDH), were more frequently reported in Lango than in Busoga, and they were associated with significantly lower agreement levels. In Uganda, single test line RDTs are the most commonly used, making them more familiar to HCWs, especially in rural districts. However, in 2020, molecular surveillance studies in Uganda provided evidence of HRP2/3 deletions in certain regions, which prompted a policy decision to provide RDTs that included pLDH in the affected areas [27, 28]. Newly introduced two test-line RDTs, such as Bioline Malaria Pf/Pan and Bioline Malaria Pf (HRP2/pLDH), require additional training to ensure accurate interpretation. Since the introduction of these tests was not universal across study facilities, it is likely that HCWs did not receive adequate training, leading to misinterpretations. This finding highlights the importance of ensuring that HCWs receive proper training and updated reference materials for newly introduced RDTs, particularly those with different characteristics. Strengthening HCW capacity and standardising RDT use will enhance data quality, ensuring reliable malaria surveillance and case management across Uganda.

Results misrecorded as positive were more common than results misrecorded as negative, suggesting a possible systematic bias or set of factors influencing the direction of recording errors. When an HCW is in doubt about how to interpret an RDT result or lacks the expertise to confidently explain the cause of fever in the presence of a negative test, they would rather err on the side of diagnosing malaria to minimise the risk of missing a potentially life-threatening malaria diagnosis. This partly explains the higher prevalence of non-compliance with negative than positive RDT results, as observed in multiple studies in the region [29–31]. Though uncommon, results misrecorded as negative could be partially explained by transcription errors introduced in the health facility register during the recording of test results. Moreover, in this study, about 70% of RDT results were recorded in the health facility register by a staff member different from the one who performed the test. This has serious implications for the quality of OPD register data, regardless of whether it was intentional or not. Innovative solutions such as validation of malaria RDT results could be considered, especially in the Lango region, to improve data quality [32]. The other highly plausible explanation for these results could be premature interpretation of RDTs, potentially attributed to the overwhelmingly large patient numbers. This was especially common in the Lango region, which contributed the majority of RDTs observed in this study. To cope with overwhelming patient numbers, HCWs processing RDTs in batches was observed. While batched RDTs were clearly labelled with patient identification information to ensure patients were assigned correct results, this practice could have led to some RDTs being interpreted prematurely, before the recommended manufacturer test time, affecting the accuracy of the final result. Due to the unavoidable time lag between HCW’s interpretation and image capture in such scenarios, some tests initially interpreted as negative could have finished processing, indicating a positive at the time of image capture, resulting in discrepant interpretations by the external panel. The appearance of the test line on an RDT is a time-dependent function. Therefore, premature interpretation of an RDT as negative due to non-appearance of the test line could easily turn out to be a result misrecorded as negative [33].

Incorrect interpretation of RDTs and subsequent misdiagnosis of malaria may lead to inappropriate patient treatment. Results misinterpreted as negative lead to under-diagnosis and consequently under-treatment of malaria, with the danger of allowing the progression of illness to severe malaria, a life-threatening condition [18, 34, 35]. Patients incorrectly diagnosed with malaria lead to the wastage of antimalarials and may contribute to the mismanagement of nonmalarial causes of fever, potentially contributing to poor outcomes [36, 37]. Furthermore, incorrect records may result in under- or overestimation of malaria cases, affecting resource allocations and policy decisions in affected regions. This further supports the need to consider innovative strategies, such as RDT results validation, to improve the accuracy of recorded data. In Nigeria, where this approach has been tested, it has demonstrated potential to significantly improve the reliability of reported results [32].

Strengths and limitations

This study benefited from a large sample size with over 37,000 individual RDT observations. Unlike previous evaluations using digital RDT readers that have compared TPRs between health facility registers and a subset of patients whose results were read by an RDT reader, this study conducted comparisons at the individual RDT cassette level. This approach reduced potential bias associated with sampling. Nonetheless, the study had some limitations. First, RDT results recorded in laboratory registers were not captured, limiting the ability to assess whether these entries were more accurate than those in the facility register. While including these data could have enhanced understanding of within-facility consistency, it is unlikely to have affected the overall conclusions. Second, the external panel interpreted RDT results based on images captured by a smartphone camera. This interpretation could have been affected by the quality of the images submitted, limiting visibility of particular details such as faint lines. However, the study used robust quality control procedures for image interpretation, which minimised the likelihood of such errors. Third, the daily presence of RAs taking photos of RDTs over six months may have influenced HCW behaviour, potentially leading to an underestimation of misrecording rates. During the launch of the study at each site, HCWs were briefed about the study procedures, including the daily presence of the RAs and what the roles and expectations of the RAs and HCWs would be. Consensus was sought on how RAs would fit in the workplace without overly disrupting routine workflow. Additionally, although some evidence suggests that misrecording of RDT results, particularly negative results, may be linked to over-treatment and could be intentional, the study did not establish whether these misrecordings were due to deliberate actions, unintentional errors or misinterpretation of the results. Finally, when comparing these findings with those from the previously published multi-country overview [19], it is important to recognise that the two analyses employed different methodological approaches. The overview paper used multinomial logistic regression to identify factors associated with different types of inaccuracies (RDTs misrecorded as positive versus misrecorded as negative), whereas this paper applied Cohen’s kappa to measure overall agreement. Cohen’s kappa summarises overall agreement adjusted for chance, whereas multinomial logistic regression distinguishes covariates influencing false-positive and false-negative misrecordings. A variable may therefore influence the odds of specific disagreement types without significantly altering the overall kappa, which explains the variation in the factors identified as significant across the two papers.

Conclusion

The level of agreement between RDT results recorded by HCWs and those interpreted by an external panel was strong, suggesting that RDT results reported from health facilities are generally accurate. However, important proportions of RDTs were misrecorded as positive or negative, particularly in the Lango region, where factors such as high patient volume, limited HCW staffing, and the use of newer two test-line RDTs were more common. To reduce workload and minimise reporting errors, increasing the number of trained HCWs, especially in high-burden districts, should be prioritised. In addition, measures are needed to limit the number of RDT products in the market, while standardised training and reference materials should be provided when introducing new RDT products.

Finally, innovative strategies, such as validation mechanisms for RDT results, should be explored to enhance data quality, particularly in settings with high patient volumes.

Abbreviations

CI

Confidence interval

HC

Health centre

HCW

Healthcare worker

MOH

Ministry of Health

OPD

Outpatient department

Pf

Plasmodium falciparum

PfPR_2–10

Plasmodium falciparum parasite rate in the population aged 2–10 years

PR

Prevalence

RA

Research Assistant

RDT

Rapid diagnostic test

TPR

Test positivity rate

WHO

World Health Organization

Author contributions

KAL, MH and KG conceived and designed the evaluation. NS, JFN, AK, JM, NM, AM, and BA oversaw data collection activities. SC oversaw the development of the HealthPulse application used in the study. NS and KAL drafted the manuscript. JJA, RS and KAL analysed the data. NS, JFN, AK, JM, NM, MH, KG, JJA, RS, SC, BA, JO, KAL and AM critically reviewed the manuscript. All authors read and approved the final manuscript.

Funding

This evaluation was co-funded by PMI Insights and the Bill & Melinda Gates Foundation (Investment # INV-043942). PMI Insights was the global operational research and program evaluation project of the U.S. President’s Malaria Initiative (PMI). Funding for this evaluation was made possible by the generous support of the American people through the United States Agency for International Development (USAID) (Cooperative agreement # 7200AA20CA00031). The contents are the responsibility of the authors and do not necessarily reflect the views of USAID or the United States Government.

Data availability

The datasets generated and/or analysed during the current study can be provided by the corresponding author upon reasonable request.

Declarations

Ethical approval and consent to participate

Ethical approval was obtained from the Vector Control Division Research and Ethics Committee (Ref # VCDREC172) and the Uganda National Council of Science and Technology (Ref # HS2747ES) in Uganda, and the WGC IRB in the USA (Ref # 20231373). All participants provided written informed consent to participate.

Competing interests

The authors declare that they have no competing interests.