Glycated haemoglobin and fasting plasma glucose tests in the screening of outpatients for diabetes and abnormal glucose regulation in Uganda: A diagnostic accuracy study.

BACKGROUND AND OBJECTIVES: To understand the utility of glycated haemoglobin (HBA1C) in screening for diabetes and Abnormal Glucose Regulation (AGR) in primary care, we compared its performance to that of the fasting plasma glucose (FPG) test.
METHODS: This was a prospective diagnostic accuracy study conducted in eastern Uganda. Patients eligible for inclusion were consecutive adults, 30-75 years, receiving care at the outpatient department of a general hospital in eastern Uganda. We determined the sensitivity, specificity and optimum cut-off points for HBA1C and FPG tests using the oral glucose tolerance test (OGTT) as a clinical reference standard.
RESULTS: A total of 1659 participants underwent FPG testing of whom 310 were also HBA1C and OGTT tested. A total of 113 tested positive for diabetes and 168 for AGR on the OGTT. At recommended cut-off points for diabetes, the HBA1C and FPG tests had comparable sensitivity [69.8% (95% CI 46.3-86.1) versus 62.6% (95% CI 41.5-79.8), respectively] and specificity [98.6% (95% CI 95.4-99.6) versus 99.4% (95% CI 98.9-99.7), respectively]. Similarly, the sensitivity of HBA1C and the FPG tests for Abnormal Glucose Regulation (AGR) at ADA cut-offs were comparable [58.9% (95% CI 46.7-70.2) vs 47.7% (95% CI 37.3-58.4), respectively]; however, the HBA1C test had lower specificity [70.7% (95% CI 65.1-75.8)] than the FPG test [93.5% (95% CI 88.6-96.4)]. At the optimum cut-offs points for diabetes [45.0 mmol/mol (6.3%) for HBA1C and 6.4 mmol/L (115.2 mg/dl) for FPG], HBA1C and FPG sensitivity [71.2% (95% CI 46.9-87.8) versus 72.7% (95% CI 49.5-87.8), respectively] and specificity [95.1% (95% CI91.8 97.2) versus 98.7% (95% CI 98.0 99.2), respectively] were comparable. Similarly, at the optimum cut-off points for AGR [42.0 mmol/mol (6.0%) for the HBA1C and 5.5 mmol/l (99.0 mg/dl) for the FPG test], HBA1C and FPG sensitivity [42.3% (95% CI 31.8-53.6) and 53.2 (95% CI 43.1-63.1), respectively] and specificity [89.1% (95% CI 84.1 92.7) and 92.7% (95% CI 91.0 94.1), respectively] were comparable. DISCUSSION: HBA1C is a viable alternative diabetes screening and confirmatory test to the FPG test; however, the utility of both tests in screening for prediabetes in this outpatient population is limited.

Introduction

Type 2 diabetes (T2D) is one of the leading public health challenges today. About three out of four (79%) of all people with diabetes live in low- and middle-income countries [1]. Sub-Saharan Africa (SSA) is the region with the highest proportion of undiagnosed diabetes globally; of the 19.4 million people living with diabetes in SSA, 11.6 million (59.7%) are unaware of their condition [2]. Diabetes has been associated with different comorbidities including severe bacterial infections, such as tuberculosis, and viral infections including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [3]. As the world faces an unprecedented SARS-CoV-2 pandemic, improving diabetes diagnosis could not be overemphasized [4].

Individuals with undiagnosed diabetes develop uncontrolled hyperglycemia which puts them at high risk of developing micro- and macrovascular T2D diabetes complications. The diagnostic gaps in related to T2D seen in SSA are due to a low awareness of T2D risk factors and poor access to diabetes testing facilities among the general population, limited knowledge of screening and diagnosis among health professionals, and limited access to low cost screening tests [5, 6].

The two tests commonly used for T2D testing are the Fasting Plasma Glucose (FPG) and the Glycated Haemoglobin test (HBA1C). The HBA1C test has a major practical advantage over the FPG test in that it does not require overnight fasting. Because of this, test results can be availed to the patient at their index health facility visit. The HBA1C test also measures the average glycaemia over the preceding 2–3 months [7], unlike the FPG test which measures glycemia at a point in time. It also has less intra-individual variability [8], greater analytical stability, and is a better predictor of microvascular disease than FPG [9].

However, HBA1C levels vary by race [10, 11], but no such racial variation in FPG levels has been reported. In addition, red blood cell lifespan determines the duration of exposure to glucose. Therefore, conditions that increase red cell turnover–such as haemoglobinopathies, malaria, and haemolytic anaemias–are associated with spuriously low HBA1C results [12]. Other conditions, such as iron deficiency anaemia (IDA), on the contrary, are associated with elevated HBA1C levels [13, 14].

Moreover, the WHO-recommended HBA1C cut-offs have been derived from predominantly non-black populations. It is not clear whether these figures could be extrapolated to geographical areas with higher prevalence of heamoglobinapathies, malaria and anaemia such as SSA [15], like Uganda. Sickle cell disease, malaria and other parasitic infections are endemic, with a third of the women aged 15–49 years and 16% of the men anaemic [16]. Validating screening tests for use in sub-Saharan Africa is crucial for tackling the rising type 2 diabetes epidemic in the region. In this study, we compared the diagnostic performance of point-of-care (POC) HBA1C and FPG tests when used to screen for diabetes and prediabetes in an outpatient population.

Methods

Setting

The study was conducted at the outpatient department of Iganga general hospital in eastern Uganda, 120 kilometres from Kampala, the capital city of Uganda. The hospital serves communities in Iganga and surrounding districts, whose main economic activities are subsistence farming, fishing, and petty trading. In addition to inpatient services, the hospital operates a busy outpatient department (OPD) for patients referred from lower-level health facilities; however, most patients self-refer to the hospital. The outpatient department provides general preventive, promotional, curative, and maternity services. Patients suspected of having diabetes in the OPD are referred to the diabetes clinic, which is conducted every Tuesday.

Study design

A diagnostic accuracy study was conducted. Eligible participants were patients aged 30–75 years accessing care at the OPD of Iganga hospital. Individuals who were known to have diabetes and on anti-hyperglycaemic drugs, anti-psychotic drugs, systemic steroids, those known to have sickle cell disease or with clinical features suggestive of sickle cell disease, and those with a history of having undergone blood transfusion within the previous 3 months were excluded.

A sample of 110 participants with diabetes was required to demonstrate noninferiority of the HBA1C test compared to the FPG for a minimum relative sensitivity of 0.75 for the two tests with a power of 80% and a significance level of 0.05, assuming a sensitivity of 78% and 81% for the HBA1C and the FPG test [17], respectively.

We used a two-stage sampling strategy. The first stage involved consecutive sampling and FPG testing of all eligible study participants. At the second stage, test result-based sampling [18] was conducted such that only a sub-sample of the study participants underwent both the HBA1C test and verification with a two-hour oral glucose tolerance test (OGTT) clinical reference standard. This sampling was done among individuals with more than 6.0 mmol/l and those with less than or equal to 6.0 mmol/l in a 1:1 ratio. The two tests were conducted in every participant who scored greater than 6.0 mmol/l on the first stage FPG test. In addition, the next study participant who scored 6.0 mmol/l or less on the first stage FPG test was also HBA1C and OGTT tested to obtain the 1:1 ratio (). The approach ensured that the sample of individuals undergoing all three tests had higher proportion of individuals with hyperglycaemia and diabetes. This way, the sample size could be attained at a minimal cost. The two-hour OGTT clinical reference was chosen because it is is a more sensitive than either the FPG or HBA1C tests and better at mimicing postprandial hyperglycemia, the earliest form of dysglycemia [19, 20].

Data collection procedures

The data collection forms were programmed using Open Data Kit (ODK) platform. Data was collected by research assistants (nurses and laboratory technologists) experienced in diabetes screening who used android-based tablets. All data entered on the tablet was transmitted to a secure server via internet. The study participants were recruited from the waiting area of the outpatient department. Eligible study participants were requested to return to the hospital the next morning having had their last meal at dinnertime.

A structured questionnaire was used to collect sociodemographic data such as on age, sex, and education level. Anthropometric measurements were made following standard procedures. Height was measured to the nearest centimetre using seca® 213 stadiometers (seca gmbh, Germany), with the participant standing upright. Weight was measured to the nearest 0.1 kg using a seca® 813 (seca gmbh, Germany) weighing scale, with the participant wearing light clothing without shoes. Waist circumference was measured in centimetres using a tape measure strapped along a horizontal plane at the midpoint between the bottom of the patient’s last rib and the top of the iliac crest. Body mass index (BMI) was calculated using the formula, weight in kg/(Height in m)2. The blood pressure was determined from the mean of three readings measured 10 minutes apart using an Omron M2 blood pressure monitor (model HEM-7121-E, Omron Corporation) with the patient seated.

Laboratory investigations

In all cases, the FPG and HBA1C were conducted by the same research assistants before the OGTT. The FPG test was performed on each consenting adult from a drop of capillary blood drawn from a finger prick using a spring-loaded lancet. The blood was analysed using the handheld Accu-Chek® Active blood glucose meter (Roche Diagnostics GmbH, Mannheim, Germany), which is readily available on the Ugandan market. This glucometer meets ISO 15197:2013 specifications, according to the manufacturer’s product insert. This means that at least 95% of the results of this glucometer are within ±15 mg/dl at glucose concentrations <100 mg/dl and within ±15% at ≥100 mg/dl compared to a traceable laboratory method [21]. For the OGTT, participants drank 75 g of anhydrous glucose dissolved in 250 mls of water after which they were asked to sit on a bench within the testing area. After 2 hours, a capillary blood sample was taken by finger prick and tested with the glucometer. Similarly, the HBA1C test was done on capillary blood from a finger prick using a Cobas b 101 system (Roche Diagnostics GmbH, Mannheim, Germany), a benchtop device accredited by the National Glycohemoglobin Standardization Program (NGSP) and standardized to the Diabetes Control and Complications Trial (DCCT) assay. The coefficient of variation of the Cobas b 101 instrument is less than 5% [22], as recommended by the International Federation of Clinical Chemistry and Laboratory Medicine for HBA1C POC tests [23, 24].

Definition of diabetes and abnormal glucose regulation

The Standards for Reporting of Diagnostic Accuracy Studies (STARD) 2015 checklist [25] was followed with FPG and HBA1C as index tests and OGTT as the reference standard. Diabetes was defined based on OGTT ≥11.1 mmol/L, FPG ≥7.0 mmol/L, and HBA1C ≥48 mmol/mol, per World Health Organization (WHO) [26] and the American Diabetes Association (ADA) criteria (ADA 2021). The term ’Abnormal Glucose Regulation’ (AGR) was used to define a composite of prediabetes and diabetes based on the following criteria: OGTT ≥ 7.8 mmol/L (WHO and ADA criteria), FPG ≥6.1 mmol/L (WHO criteria) and HBA1C ≥ 39 mmol/mol (ADA criteria).

Statistical analysis

Continuous variables were presented as mean and categorical variables were presented as frequencies and proportions. Subjects were divided into 3 groups based on whether they had normal glycaemia, AGR or diabetes, according to OGTT criteria.

The diagnostic accuracy of HBA1C and FPG to diagnose diabetes and abnormal glucose regulation when OGTT was used as the reference standard was assessed on seven dimensions: sensitivity, specificity, positive likelihood ratio and negative likelihood ratio, positive predictive value (PPV), negative predictive value (NPV) and interval likelihood ratios. These were computed after verification bias-correction of test results using inverse probability weighting (Pepe 2003) (See S1–S4 Appendices). We explored how the sensitivity and specificity of the FPG and HBA1C would vary in a simultaneous testing using a scenario tree.

Receiver operating characteristic (ROC) curves were plotted using Stata version 14 and the area under curves (AUC) for both the HBA1C and FPG tests determined using the R statistical package. P-values ≤ 0.05 were considered statistically significant. We considered the optimum cut-offs for the two tests as the points on their ROC curves where sensitivity and specificity were highest.

Ethical considerations

The study was approved by Makerere University School of Public Health Higher Degrees, Research and Ethics Committee and the Uganda National Council of Science and Technology (reference number: HS 2611). Administrative clearance was sought and received from Iganga Hospital. All participants provided written informed consent. To ensure confidentiality, anonymous identifiers were used during data collection, all tablets used for data collection were password protected, and data was encrypted during transmission to prevent unauthorized access. All study participants diagnosed with diabetes during the study were referred to the Iganga Hospital diabetes clinic for further management.

Results

The characteristics of the participants who underwent testing with both index tests and the clinical reference standard are summarized in . The mean age was 49.0 years (95% CI 47.8–50.2). The mean BMI was 26.8 kg/m2 (95% CI 26.2–27.4). Women had a higher mean BMI [27.4 kg/m2 (95% CI 26.7–28.1)] compared to men [24.8 kg/m2 (95% CI 23.5–25.9)]. Similarly, a higher proportion of women [27.4% (95% CI 26.7–28.1] reported eating vegetables and fruits daily compared to men [24.8% (95% CI 23.5–25.9)] ().

The Standards for Reporting of Diagnostic Accuracy Studies flow diagram is shown in , 2113 subjects were screened for study eligibility between 6 January 2020 and 26 February 2021. Of these, 1659 participants consented to participate in the study and underwent an initial FPG test. Those whose FPG results were greater than 6.0 mmol/l (155/1659) underwent HBA1C and OGTT testing and 110 tested positive for diabetes on the OGTT (). Similarly, 155 of those whose FPG results were less or equal to 6.0 mmol/l were also HBA1C and OGTT tested and 3 tested positive for diabetes on the OGTT.

Sensitivity and sensitivity as screening tests

The sensitivity of the FPG and HBA1C tests when screening for diabetes based on WHO and ADA criteria [69.8% (95% CI 46.3–86.1)] versus 62.6% (95% CI 41.5–79.8), respectively] were comparable and moderate (). Similarly, the specificity of the FPG and HBA1C tests were not only comparable [99.4% (95% CI 98.9–99.7) versus 98.6% (95% CI 95.4–99.6), respectively] but high. The positive predictive value (PPV) of the FPG test (0.91) was substantially higher than that of HBA1C test (0.82). The sensitivity of the HBA1C test when screening for AGR according to ADA criteria [58.9% (95% CI 46.7%-70.2%)] was only moderate, and was comparable to that of the FPG test [47.7% (95% CI 37.3%-58.4%)]. However, the specificity of the FPG test in detecting AGR [93.5% (95% CI 88.6%-96.4%)] was substantially greater than that of the HBA1C test [70.7% (95% CI 65.1%-75.8%)] (see Table 2). The sensitivity and specificity of the FPG test in detecting AGR according to the WHO criteria were 29.0% (95% CI 22.5%-36.6%) and 98.3% (95% CI 97.4%-98.9%), respectively.

Table 2

Sensitivity, specificity, PPV and NPV of the FPG and HBA1C tests in the diagnosis of diabetes and AGR at the OGTT cut-offs recommended by ADA and WHO.

	Test	Sensitivity, %		Specificity, %		PPV		NPV
		(95% CI)		(95% CI)
Diabetes	FPG	62.6 (41.5–79.8)		99.4 (98.9–99.7)		0.91		0.97
	HBA 1C	69.8 (46.3, 86.1)		98.6 (95.4–99.6)		0.82		0.97
AGR		Based on ADA criteria	Based on WHO criteria	Based on ADA criteria	Based on WHO criteria	Based on ADA criteria	Based on WHO criteria	Based on ADA criteria	Based on WHO criteria
	FPG	47.7 (37.3–58.4)	29.0 (22.5–36.6)	93.5 (88.6–96.4)	98.3 (97.4–98.9)	0.74	0.87	0.82	0.78
	HBA 1C	58.9 (46.7–70.2)	NA	70.7 (65.1–75.8)	NA	0.45	NA	0.82	NA

Optimal cut-off values

Fig 2 shows the ROC curves for FPG and HBA1C in screening for Diabetes and AGR. The FPG and HBA1C tests had comparable discriminatory capacity for both diabetes (AUC = 0.85 versus 0.86, respectively) and AGR (AUC = 0.74 versus 0.68, respectively).

Fig 2

ROC curves for FPG and HBA1C in the diagnosis of diabetes and AGR.

The optimum diabetes cut-offs for the HBA and FPG for diabetes were 45 mmol/mol (6.3%) and 6.40 mmol/L (115 mg/dL) (); the sensitivities and specificities for the two tests at these cut-off points (71.2% and 98.7%, respectively, for the HBA test and 72.7% and 98.7%, respectively, for the FPG) were comparable. On the other hand, the optimum AGR cut-offs for the HBA and the FPG test were 42.0 mmol/mol (6.0%) and 5.50 mmol/L (99 mg/dL), respectively. The sensitivity and specificity for the HBA test (42.3% and 89.1%, respectively) and FPG (53.2% and 92.7%, respectively) at these points were similarly comparable.

Discussion

The question of whether the HBA1C test performs adequately compared to the FPG test when used to screen for diabetes and prediabetes in SSA is fundamental to its wider uptake in the region [27]. In this validity study, which was conducted at a general hospital in eastern Uganda, generally the performance of the HBA1C test in screening for diabetes and AGR was comparable to that of the FPG test. However, the specificity of the HBA1C test in screening for AGR at ADA recommended cut-offs was lower than that of the FPG. This is one of the few FPG and HBA1C validity studies conducted among a SSA outpatient population that used an OGTT clinical reference standard.

The sensitivity of the HBA1C when screening for diabetes was moderate at standard cut-off points. Hird et al [28] found a similar HBA1C diabetes sensitivity of 70.3% among a black South African general population in the Durban Diabetes Study. However, an earlier study conducted among a mixed ancestry South African population, the Bellville-South cohort [29], found an even lower HBA1C sensitivity estimate for diabetes (45.9%). The disparity in HBA1C sensitivity evident in these studies could be explained by differences in patient spectrum affecting both test accuracy and prevalence [30]. The mixed ancestry population had a much higher diabetes prevalence (46.9%) [29] compared to the black South African population (12.9%) [28] and our study population (8.4%). Besides, our study was conducted among outpatients, a population likely to have major clinical differences compared to the general populations that were the focus of the other two studies. The sensitivity of the HBA1C test was higher than that of the FPG test in this study which points to the potential for the HBA1C test to correct detect a higher proportion of individuals with diabetes than the FPG test for an outpatient population in this setting at standard cut-off points. Used together in a simultaneous testing scenario, the two tests could have a combined sensitivity as high as 88.7% at standard cut-off points. However, testing individuals with both tests in this way would require considerable resources and may only be possible in some cases.

The HBA1C test had a high specificity at standard cut-off points (98.4%), which was comparable to that of the FPG test (99.4%). This finding is consistent with that of Zemlin et al [29] and Hird et al [28] who found HBA1C specificities of 96% and 98.7%, respectively. The high specificity of both the HBA1C and FPG tests highlights the minimal number of individuals with false positive diabetes diagnoses detected by either test. This finding supports the utility of the HBA1C test, alongside the FPG test, as confirmatory tests.

At the optimum cut-off point, 45.0 mmol/mol (6.3%), which was lower than the WHO and ADA recommended threshold, the HBA1C test had a specificity of 95.1%. This threshold was higher than that reported by Zemlin et al [29] and Hird et al [28] (77% and 92%, respectively) from South Africa who found lower cut-off points [43 mmol/mol (6.1%) and (≥42 mmol/mol [6.0%]), respectively]; however, the two studies reported higher sensitivities (77% and 92%, respectively) compared to our study (71.2%). These differences may be attributed to the patient spectrum differences discussed in the previous paragraph for the three studies at standard diabetes cut-off points. Also notable is that the optimum FPG cut-off point [6.4 mmol/L (115 mg/dl)] was also lower than the WHO and ADA recommended threshold. The sensitivity (72.7%) and specificity (98.7%) of the FPG test at this point were comparable to those of the HBA1C test. The HBA1C and FPG tests could potentially be better at detecting diabetes at their optimal cut-off points than the currently recommended cut-offs.

The FPG test had peak AGR sensitivity (53.2%) and a high specificity (98.7%) at its optimal cut-off point [5.5 mmol/L (99 mg/dl)]. This FPG cut-off point was lower than the threshold recommended by the the WHO [6.1 mmol/L (110 mg/dl)] but close to the ADA cut-point [5.6 mmol/L (100 mg/dl)]. Compared to the WHO recommended cut-off point, such a low FPG cut-off point could lead to overdiagnosis of AGR with the potential to overwhelm the under-resourced public health system [31]. For the HBA1C test, the optimum cut-off point [HBA1C ≥ 42 mmol/mol (6.0%)] was higher than the ADA recommended cut-off point [HBA1C ≥ 39 mmol/mol (5.6%)] and similar to the threshold recommended by the International Expert Committee [32]. Therefore, fewer people would be correctly diagnosed with AGR at the optimum HBA1C cut-off compared to the number that would be at the HBA1C ADA cut-off point. Using the two tests together in a simultaneous testing scenario for AGR at the WHO cut-off for the FPG test and the optimum cut-off for the HBA1C test would lead to only a slightly higher combined sensitivity of 59%. However, the combined sensitivity would be much higher (81%) for a simultaneous testing strategy involving the FPG test at its optimum cut-off point and the HBA1C at the ADA-recommended cut-off point.

The HBA1C test seems like a viable alternative to the FPG test for the screening of type 2 diabetes among the outpatient population, based on the findings of this study. The HBA1C test is particularly suited for screening in the outpatient population as it is minimally affected by acute illness [33] or short-term physical exertion, as may occur when patients walk or ride to the health facility. As it does not require fasting, it would be available to patients throughout the day and night. To improve the yield and positive predictive value of the test, the HBA1C test could be targeted to high-risk individuals. Obesity, physical inactivity and poor dietary diversity are some of the known diabetes risk factors in the general population in eastern Uganda [34]. A growing body of evidence points to differences in the risk profiles of diabetes in SSA compared to populations elsewhere; for example, the early onset of diabetes in SSA and the large burden of disease among individuals with diabetes with normal BMI [34, 35]. Optimal diabetes screening strategies for Uganda and similar settings should be based on a better understanding of the appropriate diabetes risk strata to ensure efficiency. We found the HBA1C to have the potential to correctly detect more individuals with diabetes and prediabetes in this population. Since this test is more costly than the FPG test any screening strategy based on the HBA1C test would call for further studies to compare its cost-effectiveness to that of the FPG test.

HBA1C POC testing is a viable alternative to FPG testing in the screening of diabetes in the outpatient setting. Both tests had high specificity for diabetes and could therefore be used for second-line confirmatory screening of individuals who test positive. The sensitivity for both tests for testing for AGR and therefore prediabetes are low, limiting their utility in the screening of prediabetes.

Recommendations

Our study has implications on the testing of diabetes in Uganda and similar contexts. Providing HBA1C testing to complement or as an alternative to FPG screening at the health facility level has the potential to improve the detection of diabetes. The FPG test could be provided to patients who arrive at the facility early and are still in a fasted state, reserving the HBA1C test for those who arrive later in the day. Either test could be used for confirmatory testing for diabetes. However, confirmatory testing for prediabetes should be reserved for the FPG at the WHO cut-off point. Where resources allow, a combination of the HBA1C test and the FPG tests could be used to screen both diabetes and prediabetes. Future studies should evaluate the performance of the HBA1C and FPG in sub-populations with high red cell turnover, and their effectiveness and cost effectiveness in the targeted screening of diabetes and prediabetes among high risk populations.

Study limitations

Previous research has shown that HIV patients on protease inhibitors and nucleoside reverse transcriptase inhibitors have a different diabetes risk profile from the general population, our study was underpowered to evaluate the performance of the HBA1C in this patient population. Even so, previous research in the Uganda population [36] found that the FPG and HBA1C tests correlation among HIV patients was similar to that among members of the general population (r  =  0.69 vs. r  =  0.66). Similarly, there was no evidence of a statistically difference between mean FPG and HBA1C levels between the two groups. However, a study conducted in a population of HIV patients on antiretroviral therapy in South Africa [37] found a very low sensitivity of 37% at the WHO recommended cut-of.

We did not evaluate HBA1C performance among individuals with haemoglobin variants, such as sickle cell trait, which could interfere lead to diabetes misclassification [38]. However, the Cobas b101 benchtop analyzer used in our study is stable to sickle cell trait interference [23] and would be expected to perform to the same level in this patient category. We were not able able to assess the validity of the HBA1C test for sub-populations with anaemia and other conditions leading to high red blood cell turnover, as our study was not powered for this analysis. Similarly, we did not collect data on the pregnancy status of the female participants of this study which is another limitation of our findings. Finally, the two-hour OGTT reference only measures one of several dysglycaemic aetiologies, limiting its utility as a diabetes and prediabetes clinical reference standard.

(DOCX)

Click here for additional data file.

STARD for abstracts: Essential items for reporting diagnostic accuracy studies in journal or conference abstracts.

(DOCX)

Click here for additional data file.

Calculations for the sensitivity and specificity of the FPG and HBA1C tests when used to screen for diabetes.

(DOCX)

Click here for additional data file.

Calculations for the sensitivity and specificity of the FPG and HBA1C tests when used to screen for AGR.

(DOCX)

Click here for additional data file.

Computation of sensitivity and specificity for optimum cut-offs for Diabetes and the corresponding 95% CIs.

(DOCX)

Click here for additional data file.

ROC and interval likelihood ratios for the FPG and HBA1C tests when used to screen for diabetes and AGR.

(DOCX)

Click here for additional data file.

Transfer Alert

This paper was transferred from another journal. As a result, its full editorial history (including decision letters, peer reviews and author responses) may not be present.

25 May 2022

25 Jun 2022

EDITORIAL COMMENTS

An interesting and relevant work to inform a key research gap.

The manuscript will benefit more by addressing the following:

Lin 205 to 208 authors will need to reference the table which describes these results. It is not clear as it is 155 participants are those with above 6mmol/L or "above or less."

Response: The 155 participants are those whose Fasting Plasma Glucose (FPG) results were greater than 6 mmol/L. We tested an equal number of participants whose results were less than or equal to 6 mmol/L. As advised, we have edited the manuscript to make this clearer and referenced Fig 1, which illustrates those results. These changes are on lines 239-243 of the manuscript’s marked-up version.

In the result section authors have described findings of the outcomes of the study before description of demographics and clinical characteristics. Authors will need to consistently follow the STARD guideline and use the STARD checklist to REWRITE their manuscript.

Response: We have corrected this error in the revised version of the protocol, moving the section at the beginning of the study outcomes from line 220 to 237. In this manuscript version, the results section begins with a description of the demographic and clinical characteristics. We have also rewritten the various sections of the manuscript according to STARD guidelines and completed STARD checklists for both the abstract and the body of the manuscript (see Supporting file 4 and Supporting file 5). The corresponding revisions made to the manuscript are as follows:

For the abstract, we have revised the subsections and content to reflect current recommendations in the “STARD for abstracts”. We have added a “Background and objectives” section (line 30 in the version of the manuscript with tracked changes) and renamed the “Conclusion” section “Discussion” (line 64). In the methods, we have identified the study as prospective (line 35), added the eligibility criteria (lines 36-37), noted that sampling was consecutive (line 36) and identified Glycated haemoglobin test (HBA1C) and FPG as index tests and Oral Glucose Tolerance Test (OGTT) as the clinical reference standard. In the results, we mention the number of participants with diabetes and those with Abnormal Glucose Regulation (AGR) included in the analysis (line 42).

In the body of the manuscript, we have added the rationale for choosing OGTT as the clinical reference standard (lines 139-141). We have also explained that the index tests (FPG and HBA1C) were conducted before the OGTT in all cases (lines 162-163). In addition, we explained how we explored the sensitivity and specificity of the FPG and HBA1C would vary in a simultaneous testing scenario (lines 202-204). We moved the STARDS diagram to the results section (line 244) and updated it with the information on the number of participants with AGR according to the clinical reference standard.

We have also aligned the manuscript to the terminologies recommended by the STARD guidelines. Throughout the manuscript, we have replaced the word “gold standard” with “clinical reference standard” and referred to the HBA1C and FPG tests as “index tests”.

In table 2, authors are reporting comparable specificity (overlapping CI) however, in the result description and summary in the discussion, authors have reported low specificity of HBA1c as low specificity. Authors will need to clarify or rectify.

Indeed, the previous version of the manuscript had several inconsistencies regarding the interpretation of overlapping CIs. We have corrected this, noting that HBA1C sensitivity and specificity were comparable at both recommended and optimum cut-off points for diabetes. This is also true for AGR except for specificity at ADA recommended cut-off points, where it was lower for the HBA1C test [70.7% (95% CI 65.1-75.8)] and non-overlapping to that of the FPG test [93.5% (95% CI 88.6-96.4)]. These changes are on lines 43-58, 273, 280-283.

REVIEWER COMMENTS

Reviewer #1: The investigators reported in their manuscript that had applied 2-stage sampling strategies. However, in both cases, the approaches were non-probability sampling in nature. Strangely, the inferences were heavily dependent on probability sampling analysis. For all practical purposes, the so-called "consecutive sampling" is a non-probability sampling method. One of the immediate impacts (despite all posteriori statistical manoeuvres done) of such 'design error' include estimates that blow out disproportionately out of the bounded confidence interval values. That can be evident in their manuscript since optimum level specificity for their HbA1c had an estimate (98.7%) outside the stated 95% C.I. (96.2% - 98.4%) in both abstracts (refer to line 47 of page 2) and even in their table 3.

Usually, this thinking invalidates the entire analysis of their primary analysis.

However, I am optimistic about the findings, based on both 'near-infinitely large sample size', as well as 'inverse probability weighting', that controlled for bias in the primary findings (sensitivity, specificity, PPV & NPV), obtained out of likely 'partial verification bias' prominent by design. It is highly likely, that their large sample size approximate the 'real population' of interest and therefore, somehow offsets all the doubted possibilities of significant biases. A sensitivity analysis is likely warranted in this scenario.

Response: Our decision is to use a consecutive was informed by the desire to enrol every eligible participant receiving care at the outpatient department. As such, our sample was highly representative of the accessible population [1]. In addition, the study was conducted for several months, ensuring that it is representative of the target population, i.e. patients accessible to ambulatory care at a district hospital. We, therefore, believe that our probability-based inferences and estimates are valid.

Thank you for pointing out the out-of-range FPG specificity point estimate. Upon review, we realise that we used the NPV instead of the specificity to compute the corresponding 95% CIs. We have corrected this in the current version. The new 95% CI is 98.0-99.2. We have revised the current version of the manuscript (lines 54 and 276 of the tracked changes version) accordingly

We appreciate your observations regarding the large sample size as a strength for this study.

2. The study design was better suited as a population-based study instead of a hospital-based analysis.

2.1 - as stated by investigators, Iganga hospital is a referral health facility; hence, the studied sample is unlikely to represent the entire population.

There is a possibility that their findings are mere 'hospital statistics' data. Time after time, hospital statistics data are known to be 'negatively biased' with respect to the clinical diagnostic utility studies. No wonder, their sample has an age range of 30-75 years. Highly likely those 'older than 75 years cohort' who are the most at risk to be diabetic were selected out from participating in this study. Likewise, the 'younger cohort' (<30 years) who are otherwise, least likely to be impaired glucose tolerant/diabetic were unlikely to be part of this study

Response: Our decision to perform the study in an outpatient hospital population was a pragmatic choice motivated by our desire to inform opportunistic screening at the health facility level. This is probably a more realistic strategy than mass screening, which, in addition to being costly and out of reach for low and middle-income countries, may provide no added value regarding patients’ positive clinical outcomes. [2] It was never our intervention to make inferences about the general population. In our previous work on diabetes diagnosis, [3] we found that patients with diabetes often visit health facilities on multiple occasions in pursuit of a diagnosis but are not screened for diabetes. We, therefore, felt that evaluating test performance in an outpatient screening scenario would be a worthwhile investment to inform diabetes screening policy and practice in this setting.

Limiting study participation to individuals no younger than 30 years was meant to minimise the potential recruitment of individuals with type 1 diabetes, which tends to develop earlier compared to type 2 diabetes, the focus of this study. For the upper limit of 75 years, we purposed to exclude the very elderly, as HBA1C may be inaccurate in this age category due to comorbidities that affect red blood cell lifespan. [4]

2.2. Prevalent clinical conditions in the population that can significantly affect HbA1c levels were not reported to have been controlled in the study by design.

For instance, how many (and proportion by %) of their participants in the study were pregnant women? and how many (and proportion by %) were HIV-seropositive on ante-retroviral treatment?

Both conditions (quite prevalent in Uganda) are known to affect the glycation process significantly and with a severe impact on values of the HbA1c test in the diagnosis of diabetes mellitus.

Response: We acknowledge the lack of data on the performance of the index tests, especially HBA1C, among HIV-positive individuals as a study limitation which we acknowledged in the original version of the manuscript (lines 370-378 in the current version with tracked changes). We have acknowledged the lack of data on the pregnancy status of the female participants in this study as another study limitation (lines 385-386).

2.3. Investigators reported in their methods section that continuous variables were to be reported in "mean +/- S.D."

However, in the results, most of the continuous variables were reported in "mean with 95% C.I." Even though there is a natural mechanism for conversion from one to another, consistency is key.

I would advice they choose one and leave the other in their reporting strategies.

Response: Thanks for the observation regarding the inconsistency in reporting continuous variables. We have removed references to standard deviations (line 190 of the version with tracked changes).

2.4. Arbitrary Cut-off points: why was HbA1c level for abnormal glucose regulation (standard) made at 5.7% instead of the usual 5.6%?

- Can't that decision results to 'lower sensitivity' as stated in the manuscript?

- How different will it be if they stick to the standard HbA1c level cut-off point of 5.6%?

Response: According to the ADA, [5] the prediabetes range per the HBA1C test is 5.7-6.4%. This corresponds directly to the cut-off of ≥5.7%, which we used to compute and report HBA1C sensitivity and specificity in Table 3. According to this classification, a patient with an HBA1C score of 5.6% would be considered normoglycemic. Therefore, we believe that our results provide an accurate estimate of HBA1C performance in screening for AGR based on our data.

References

1. Roach KE. A clinician’s guide to specification and sampling. J Orthop Sports Phys Ther. 2001;31: 753–758. doi:10.2519/jospt.2001.31.12.753

2. Durao S, Ajumobi O, Kredo T, Naude C, Levitt NS, Steyn K, et al. Evidence insufficient to confirm the value of population screening for diabetes and hypertension in low- and-middle-income settings. S Afr Med J. 2015;105: 98–102. doi:10.7196/SAMJ.8819

3. Kasujja FX, Nuwaha F, Daivadanam M, Kiguli J, Etajak S, Mayega RW. Understanding the diagnostic delays and pathways for diabetes in eastern Uganda: A qualitative study. PLOS ONE. 2021;16: e0250421. doi:10.1371/journal.pone.0250421

4. LeRoith D, Halter JB. Diagnosis of Diabetes in Older Adults. Diabetes Care. 2020;43: 1373–1374. doi:10.2337/dci20-0013

5. ADA. Standards of Medical Care in Diabetes—2022 Abridged for Primary Care Providers. Clin Diabetes. 2022;40: 10–38. doi:10.2337/cd22-as01

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

21 Jul 2022

Glycated Haemoglobin and Fasting Plasma Glucose tests in the screening of outpatients for diabetes and abnormal glucose regulation in Uganda: a diagnostic accuracy study

PONE-D-22-06422R1

Dear Dr. Francis Xavier Kasujja,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Fredrick Baragi Haraka, MD, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

None

28 Jul 2022

PONE-D-22-06422R1

Glycated Haemoglobin and Fasting Plasma Glucose tests in the screening of outpatients for diabetes and abnormal glucose regulation in Uganda: a diagnostic accuracy study

Dear Dr. Kasujja:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Fredrick Baragi Haraka

Academic Editor

PLOS ONE

Table 1

Descriptive statistics.

Characteristics	Total (N = 310)	Men (N = 69)	Women (N = 241)	p-value
Age (years)	49.0 (47.8, 50.2)	51.0 (48.1, 53.8)	48.4 (47.1, 49.7)	0.08
BMI (kg/m2)*	26.8 (26.2, 27.4)	24.8 (23.5, 25.9)	27.4 (26.7, 28.1)	<0.00
Waist circumference (cm)	89.9 (88.4, 91.5)	87.1 (83.9, 90.4)	90.73 (89.0, 92.5)	0.06
Systolic blood pressure (mmHg)	133.6 (131.1, 136.1)	132.6 (127.6, 137.7)	133.9 (130.9, 136.8)	0.69
Diastolic blood pressure (mmHg)	88.0 (86.5, 89.6)	87.1 (83.7, 90.5)	88.3 (86.5, 90.1)	0.51
FPG (mmol/l)	7.4 (6.9, 7.9)	8.3 (7.0, 9.6)	7.2 (6.6, 7.7)	0.10
2-hour PG (mmol/mol)	11.4 (10.5, 12.2)	12.0 (10.1, 13.9)	11.2 (10.2, 12.2)	0.46
HBA1C (mmol/mol)	49.9 (47.2, 52.6)	53.1 (47.0, 59.2)	49.0 (46.0, 52.0)	0.21
Family history of diabetes (%)	21.0 (16.6, 25.9)	15.9 (7.3, 24.6)	22.4 (17.1, 27.7)	0.25
Engages in least 30 minutes of physical activity per day (%)	94.2 (91.0, 96.5)	92.8 (86.6, 98.9)	94.6 (91.8, 97.5)	0.56
Eats vegetables and fruits (%)*	14.8 (11.1, 19.3)	7.3 (1.1, 13.4)	17.0 (12.3, 21.8)	0.04

Data are mean (95% CI) or percentage (95% CI). Comparisons of characteristics between men and women were done using t-tests for continuous variables and proportion tests for categorical variables.

*P-value < 0.05

Table 3

Sensitivity and specificity at standard and optimum cut-off points.

Hyperglycaemic state	Test	Cut-off points‡		Sensitivity (95% CI)#		Specificity (95% CI)#
		Standard	Optimum	Standard	Optimum	Standard	Optimum
Diabetes	FPG	≥7.0 (126)	6.4 (115.2)	62.6 (41.5–79.8)	72.7 (49.5–87.8)	99.4 (98.9–99.7)	98.7 (98.0 99.2)
	HBA1C	≥48 (6.5)	45.0 (6.3)	69.8 (46.3, 86.1)	71.2 (46.5–87.6)	98.6 (95.4–99.6)	95.1 (91.8–97.2)
AGR⸸	FPG (WHO)	≥ 6.1 (110)	5.5 (99.0)	29.0 (22.5–36.6)	53.2 (43.1–63.1)	98.3 (97.4–98.9)	92.7 (91.0–94.1)
	FPG (ADA)	≥ 5.6 (100)		47.7 (37.3–58.4)		93.5 (88.6–96.4)
	HBA1C (ADA)	≥ 39 (5.7)	42.0 (6.0)	58.9 (46.7–70.2)	42.3 (31.8–53.6)	70.7 (65.1–75.8)	89.1 (84.1–92.7)

‡Units are mmol/L (mg/dl) for FPG and mmol/mol (%) for HBA1C

#Data are %