This study met the standards set by the independent MSF Ethics Review Board for retrospective analyses of routinely collected programmatic data [36]. These standards include, but are not limited to, assurances of confidentiality, involvement of local partners, and minimal harm to patients. Caregivers were informed that routine data were being collected for clinical care and could be used anonymised for programme monitoring and increasing knowledge about lead poisoning. Coded identification numbers were used, and personal identifiers and all unnecessary data were removed from the dataset. Review of anonymous, routinely collected programmatic data does not constitute research under the Nigerian National Health Research Ethics Committee guidelines. CDC involvement did not require Human Subjects Review as CDC personnel were not involved in treating patients.

The DMSA protocol in Zamfara aimed to deliver effective chelation that was acceptable to the community and appropriate to the challenging logistical context. If initial VBLL was ≥ 45 µg/dl, children were treated with oral DMSA (Chemet, Lundbeck; Succicaptal, Serb Laboratories). The contents of the DMSA capsules were mixed with Plumpy’Nut (therapeutic food paste, Nutriset) or milk for infants, or sprinkled on honey for older children. For obtunded encephalopathic children, the contents of the capsules were mixed with water, and the slurry was administered by nasogastric tube (n = 14/3,180 [0.44%] chelation courses included in analysis of end-course VBLL as a percentage of pre-course VBLL [ECP]; n = 12/2,000 [0.6%] courses excluded from ECP analysis).

Children were initially treated in June and July 2010 with 28-d DMSA 10 mg/kg twice daily (BD), based on the protocol used during the WHO- and CDC-supported intervention for a lead poisoning outbreak in Kosovo [2],[37]. Subsequently, most patients were chelated using repeated 19-d DMSA treatment schedules based on the US Food and Drug Administration's approved drug labelling. Each dose of DMSA was rounded to 100 mg (patient weight <15 kg) or 200 mg (15–24 kg) for ease of administration (Figure 1). Because of the severity of their lead intoxication, children with pre-course VBLL ≥ 120 µg/dl were treated with 10 mg/kg thrice daily (TDS) for the full 19 d. A minimum of 2 wk between courses is recommended unless blood lead levels indicate the need for more prompt treatment [35], so the VBLL measured at the end of each course determined the timing of the next course or VBLL test. All children received identical weight-based supplementation with zinc sulphate, calcium carbonate, iron sulphate, and folic acid (Table 1). Several minor revisions of the protocol that streamlined the treatment and blood test regimen and minimised burdens on patients and caretakers were undertaken following an MSF expert advisory group's interim analysis of blood lead trends during courses of chelation. Box 1 describes the revision process, Box 2 summarises key revisions, and Table 2 gives protocol details. All of the DMSA regimens have been used previously in clinical trials and clinical toxicology practice [26],[27],[38],[39],[40].

In hospital, all doses were directly observed (directly observed therapy [DOT]); thus, inpatient location is a proxy for full medication adherence and limited opportunity for further lead exposure. Village environmental remediation commenced concurrent with the first inpatient courses, so for later courses patients were hospitalised only if their village was yet to be remediated, or if they were severely ill. Once a village's environmental lead remediation was completed, non-encephalopathic children were treated as outpatients, initially with one of the two or three daily doses of DMSA directly observed during a clinic visit. Several months later, in response to a growing caseload and the request of parents, only one dose on alternate days was directly observed at the clinic (Table 2); other doses were administered by parents or guardians in the home. The reduced frequency of DOT therefore applied to all children on outpatient treatment and was due to the number of days per week that the clinic operated, not to VBLL or clinical status.

All children aged ≤5 y from contaminated villages where remediation was taking place were offered screening. Although all residents of the contaminated villages were considered to have elevated lead exposure, personnel and financial constraints resulted in the chelation protocol being targeted to children ≤5 y of age, the only subgroup in which serious morbidity and mortality had been identified. For statistical analysis, only completed courses of DMSA of 19 or 28 d started between 1 June 2010 and 30 June 2011 with valid pre- and end-course VBLLs recorded (defined below) were included (Figure 2). Except for analysis of overall VBLL decrease per child over the entire period, 5-d DMSA courses (Table 2) were excluded. Included chelation courses were administered to 1,156 children, 856 (74%) of whom were included in a report on association of VBLL with neurological features in 972 children [16].

Venepuncture was performed after cleaning the skin with soapy water to minimise lead dust contamination. VBLL was measured using the LeadCare II (Magellan) point-of-care analyser according to manufacturer protocols. For samples with lead levels >65 µg/dl, the upper quantification limit of the device, a blood dilution method utilising donor blood with lead concentration <3.0 µg/dl [41] was developed and implemented on 20 July 2010. Exclusion of courses with VBLL tests conducted before the dilution protocol was finalised did not substantially alter the results of regression modelling; thus, blood samples obtained at all points in the study period were retained. Monthly quality control was provided by the CDC using inductively coupled plasma mass spectrometry (ICPMS): point-of-care values of 120 clinical samples (including 15 samples requiring dilution in the field laboratory) were on average 4.0 µg/dl lower than ICPMS values (limits of agreement [42] [two standard deviations] −19.7 to +11.7 µg/dl).

Haemoglobin was measured with a HemoCue Hb 301 point-of-care testing system prior to 22 November 2010 and with a Sysmex KX-21N Automated Hematology Analyzer from this date. Creatinine and alanine transaminase (ALT) were measured on a HumaLyzer 2000 (Human Gesellschaft für Biochemica und Diagnostica). White blood cells (manual count prior to 22 November 2010, Sysmex KX-21N Automated Hematology Analyzer thereafter) and neutrophils (from 22 November 2010, Sysmex KX-21N Automated Hematology Analyzer) were also measured (white blood cell count was used to calculate absolute neutrophil count from percent), although less consistently because of operational constraints. A sensitivity analysis including a pre-/post-change variable in the main regression model determined that regression analyses were unchanged by the switch in analytical method for haemoglobin. Clinical data were routinely entered into an electronic database specifically designed to support patient care and programme management through automated lists of protocol-dictated actions, test result alerts, and detailed reports (Microsoft Excel 2007).

Adverse events were systematically monitored in that full blood count, creatinine, and ALT were checked regularly at the beginning and end of each course of chelation. Analysis included the completed courses of chelation in the study database. Anecdotal severe adverse events are reported here; minor clinical events (e.g., diarrhoea) were not recorded.

A complete course of DMSA was defined as >60% of days of doses being dispensed. “Pre-chelation VBLL” means first-ever VBLL (chelation naïve). “Pre-course VBLL” (“day 0”) means a VBLL obtained up to 14 d before the course commenced, used to dictate the length and dosing for an individual course. The “end-course VBLL” was obtained from ≤2 d before to ≤10 d after the end of treatment. Age at time of VBLL was categorised as 0 to <6 mo, 6 to <12 mo, 1 to <2 y, 2 to <3 y, and 3 to 5 y [43].

Haematological and biochemical parameters were categorised as follows: haemoglobin as low (<100 g/l if <2 y old; <110 g/l if 2–5 y), high (>130 g/l if <2 y;>140 g/l if 2–5 y), or normal; ALT as normal (0–42 U/l), mildly elevated (42.1–100 U/l), moderately elevated (100.1–1,000 U/l), or severely elevated (>1,000 U/l); creatinine as low (<44 µmol/l), high (>97 µmol/l), or normal; neutrophil count as normal (>1.5×10⁹/l) or as mild (1.0–1.5×10⁹/l), moderate (0.5–0.99×10⁹/l), or severe neutropenia (<0.5×10⁹/l) [44].

“Rebound” refers to the rise in VBLL that may occur in the days to weeks after a chelation-related reduction in VBLL. Rebound typically reflects the internal movement of lead stored in the bone and soft tissue compartments back into the blood following a chelation episode. However, re-exposure to external sources of lead after cessation of chelation may also influence the magnitude of rebound that is recorded.

“DMSA dose regimen” refers to the number of doses per day and number of days of each regimen, while “DMSA administration method” refers to whether the dose was administered in hospital or in the outpatient clinic/at home, and the frequency of direct observation.

We report the association between DMSA chelation and VBLL; clinical outcomes will be reported separately. The primary outcome variable was in line with that used in other studies [24],[25],[26],[27], defined here as end-course VBLL as a percentage of pre-course VBLL (ECP):

ECP = (end-course VBLL/pre-course VBLL) × 100% (1)

Thus, an ECP <100% indicates a net decline in VBLL; ECP > 100% indicates that VBLL increased. Since a key goal of chelation treatment is to reduce VBLL, a lower value of ECP is desirable. Geometric means of ECP values for the relevant subset of courses were calculated by fitting a mixed model containing only a constant term using nested random effects (village, patient) to account for repeated measurements in the same individual over multiple treatment courses. The natural logarithm of ECP was the dependant variable, and the exponential of coefficients and the 95% CIs were calculated.

Rebound was quantified as the percentage change between the end-course and subsequent VBLL where there was no administration of DMSA in the interim.

For clinical correlation, as a secondary outcome, absolute increase or decrease of end-course VBLL compared to pre-course VBLL was calculated, with arithmetic means and 95% CIs reported since these data were normally distributed, nested by village and patient.

We used mixed models using nested random effects (village, patient) [45] to assess the influence of factors on ECP and rebound, adjusting as relevant for the following: gender, age at start of each course (categorised), pre-course VBLL (for ECP), end-course VBLL (for rebound), timing of the end-course VBLL measurement relative to first DMSA dose, DMSA dose regimen, whether it was the first course of DMSA, DMSA administration method, and start-of-course haemoglobin.

In analysing the associations between treatment with a course of DMSA and ECP, it was considered important to account for variations in DMSA administration method and dose regimen. These factors were not independent, as protocols and implementation practicalities for both changed over time as the intervention progressed. In addition, the duration of TDS dosing as opposed to BD dosing was based on pre-course VBLL (Table 2), so pre-course VBLL would confound assessment of the role of dose regimen if both were included as independent variables in the model of ECP. Therefore, the primary multivariable model of interest using ECP as the outcome variable was based on 2,262 chelation courses that featured the most prevalent chelation dose regimen: 19 d of treatment (7 d TDS + 12 d BD). A secondary model assessed variations in dose regimen using (coincidentally also) 2,262 courses administered by alternate-day DOT. A subanalysis of courses commenced at VBLL ≥ 120 µg/dl on the 19-d TDS regimen assessed the impact of DMSA at very high VBLLs. Because changes in the protocol with respect to dose regimen and dose administration method were highly correlated with calendar date, the date of a course of chelation was not included as an independent variable in the models. No allowance for multiple testing was performed.

For categorical variables the largest group was used as the base reference, except for haemoglobin, where “normal” was used as the reference. Variables were added to the multivariable model if they were significant in univariate regression at p <0.1 or if they were considered clinically or programmatically important. Final multivariable models were selected using backward elimination. p-Values were calculated for the strength of association of each variable with the outcome using Wald tests. The DMSA administration method categorical variable was assessed as continuous (in order of decreasing frequency of DOT as shown in the tables) to test for trend. Interaction was assessed where plausible and when sufficient data were available in subgroups to yield precise estimates. All observations were included in the mixed regression models; courses with missing data (n = 23 [1%], primarily haemoglobin missing) were dropped by the statistical software as not being able to contribute to the model estimation. To estimate how well the variation in the data was explained, we fitted a linear regression, approximating the mixed model for 19-d single-regimen courses, by including village as a variable and adjusting for clustering by patient by using robust standard errors. Standard regression diagnostics were performed using analysis of residuals. Data were analysed using STATA 10.1 (StataCorp).

For all 3,180 courses (Table 4), ECP had a geometric mean of 74.5% (95% CI 69.7%–79.7%) (Table 5), with a mean absolute decrease of 22.3 µg/dl (95% CI 8.1–36.5) (Figure 4). At the end of a single treatment course, ECP varied from 7% to 274%. Instances in which ECP exceeded 100% occurred predominantly when the pre-course VBLL was less than 120 µg/dl (Figure 5). At the end of their most recent course, 71% of children had a VBLL lower than their pre-chelation VBLL; for most children this was an interim measure at the end of the analysis period but not the end of their treatment intervention (Figure 6).

For 882 first-ever treatment courses that met inclusion criteria, the geometric mean ECP was 68.0% (95% CI 59.2%–78.2%), with a mean absolute decrease of 29.4 µg/dl (95% CI 8.8–50.0). For 2,298 subsequent (after-first) courses, the mean ECP was 80.2% (95% CI 77.3%–83.3%), with a mean absolute decrease of 12.5 µg/dl (95% CI 10.5–14.5).

For all 19-d courses (n = 3,120) mean ECP was 75.2% (95% CI 70.5%–80.3%), with a mean absolute decrease of 22.2 µg/dl (95% CI 7.1–37.3). The lowest mean ECP (38.8% [95% CI 32.4%–46.4%]) was achieved in the 60 courses of 28-d duration administered entirely on an inpatient basis. All outpatient administration methods (fewer doses DOT) had a higher mean ECP. The mean ECP for inpatients with all doses DOT was 47.7% (95% CI 39.7%–57.3%) compared with 79.3% (95% CI 3.8%–85.2%) for outpatient courses given as alternate-day DOT (Table 5).

For 19-d courses using the 7 d TDS + 12 d BD regimen (n = 2,262) (Table 6), the final adjusted model showed ECP was significantly associated with age, pre-course VBLL, interval since previous course, administration method, timing of end-course VBLL measurement, and haemoglobin. After adjustment for the other variables included in the final model, first treatment courses were not associated with any difference in ECP compared with subsequent courses. Gender was not associated with change in ECP but was retained in the final model as a key demographic factor. ECP was 12%–17% higher in children aged <3 y than in those aged 3–5 y (Table 6). Pre-course VBLL was strongly associated with ECP: 0.54% lower ECP for every 1 µg/dl increase in pre-course VBLL within the range 45 to <120 µg/dl, with geometric mean pre-course VBLL 60.5 µg/dl (95% CI 57.2–63.2). For each extra chelation-free day since the end of the previous course, there was a 0.05% larger decrease in ECP. ECP was lowest if VBLL was measured immediately at the end of a treatment course compared to a few days after a course ended. Low haemoglobin at the start of a course was associated with a relative increase (6.7%) in ECP compared to normal haemoglobin. Linear regression including village and clustered by patient indicated that the model explained approximately 26% (R² = 0.255) of the variation in the data. Standard regression diagnostics indicated that residuals were normally distributed. There was no evidence of interaction between pre-course VBLL and administration method (p = 0.11). Administration methods that offered greater opportunity for DOT by clinic staff were associated with larger reductions in ECP during a course (test of trend p <0.001).

The relative ECP for different dose regimens was examined within the subset of courses administered by alternate-day DOT (n = 2,262). After adjustment for age, first course versus subsequent course, days since previous course ended, timing of end-course VBLL measurement, and haemoglobin (all associations similar to those in Table 6), the 19-d TDS regimen yielded an 18.6% lower ECP—and the 5 d TDS + 14 d BD regimen yielded a 4.4% lower ECP—than the 7 d TDS + 12 d BD regimen. This finding suggests that dosing three times a day for the whole course might yield the greatest reduction in VBLL. However, differences in ECP suggested by TDS versus BD dosing may instead just reflect the influence of pre-course VBLL differences (in light of the role of pre-course VBLL demonstrated in the model in Table 6). The geometric mean pre-course VBLL was 140.9 µg/dl (95% CI 134.8–147.3) for 19-d TDS courses, 58.4 µg/dl (95% CI 58.3–60.4) for 7 d TDS + 12 d BD, and 75.5 µg/dl (95% CI 66.6–85.6) for 5 d TDS + 14 d BD, relative levels that are coherent with the ECP differences.

The geometric mean ECP for 148 courses commenced at VBLL ≥ 120 µg/dl using the 19-d TDS regimen was 67.8% (95% CI 61.7%–74.4%). A mixed model confined to these courses commenced with high pre-course VBLL showed that after adjustment for age and haemoglobin (associations consistent with main model), three factors were strongly associated with ECP. A first course was associated with lower ECP than a subsequent course (−21.8%, 95% CI −34.1% to −9.5%), a finding not seen in the larger models that included initial VBLL values from 45 µg/dl. An increase in the interval between chelation courses was associated with a 0.7% per day (95% CI −1.1% to −0.3%) decrease in ECP. Every 1 µg/dl increase in pre-course VBLL in this subgroup was associated with a 0.20% (95% CI −0.33% to −0.07%) decrease in ECP.

The geometric mean ECP for 159 almost exclusively inpatient courses of 19 or 28 d (>95% of course in hospital) was 47.7% (95% CI 39.7%–57.3%) (Table 5), with a mean absolute decrease of 56.2 µg/dl (95% CI 40.7–71.6). For 145 inpatient-only courses that were also first-ever courses the decline was slightly greater, yielding an ECP of 44.9% (95% CI 35.4%–56.8%). Modelling indicated that the influence of pre-course VBLL on ECP was attenuated in the subset of patients undergoing a first course of DMSA administered entirely on an inpatient basis. In this subset (n = 145, geometric mean pre-course VBLL 106.2 µg/dl, 95% CI 87.1–129.4), after adjustment for age, for every 1 μg/dl increase in pre-course VBLL there was only a 0.09% decrease in ECP (95% CI 0.02% to 0.16%; p = 0.014).

In 2,444 pairs of tests that included a test at the end of a 19-d course of any dose regimen and the subsequent test following an interval without chelation (median 19 d, range 1–262), the geometric mean (nested by village and patient) of the VBLL rebound was 122.5% (95% CI 118.9%–126.3%), with 82% of tests showing VBLL rebound above the end-course VBLL (Figure 7). Factors significantly associated with higher rebound included higher VBLL at the start and end of the course just taken, and more chelation-free days between courses and tests. The largest percentage change was with the VBLL at the end of the course: the lower the absolute value of the end-course VBLL, the lower the percent rebound (Table 7). Exclusion of very high rebound outliers did not attenuate the strong associations in the model.

There were no anecdotal reports of severe clinical adverse events related to DMSA. One child developed pneumonia while receiving DMSA via nasogastric tube. This was diagnosed clinically on the basis of fever, hypoxia, and unilateral chest crepitations on auscultation as there were no radiology services available. The child had an AVPU (alert, voice, pain, unresponsive) [46] score of U, and the pneumonia may have been secondary to aspiration in a deeply unconscious child without sufficient gag reflex, though this was not confirmed. Alternatively, it may have represented a non-aspiration nosocomial pneumonia in a critically ill child. The child was treated with oxygen, IV ampicillin, and metronidazole and recovered. Three deaths in children with an AVPU score of P or U receiving DMSA by nasogastric tube (11% of 26 children) occurred in the first few days of treatment with severe lead poisoning as the primary cause; these patients were not included in the analysed courses as there was no end-course VBLL (Figure 3).

The incidence of severe neutropenia (neutrophil count <0.5 × 10⁹/l) after any course of DMSA was low (0.4%) and was comparable to that seen before commencement of DMSA (0% severe neutropenia pre-chelation, 0.3% severe neutropenia before any individual course). Of six children with severe neutropenia at the start of a course, three had normal neutrophil counts and three no result when the end-course VBLL was measured, but nine of the ten children with severe neutropenia at the end of a course did not have a result recorded for the start of that course. There were no recorded cases of severely elevated ALT (>1,000 U/l) either in chelation-naïve patients or after DMSA had been started. Moderately elevated ALT (100.1–1,000 U/l) was present in 1.0% of chelation-naïve children (range 4–256 U/l), increasing slightly to 1.8% after one course of DMSA, and staying constant at 2.4% after multiple courses (ALT range 3.2–498 U/l). Doubling or more of ALT during a course of DMSA occurred in 123 courses (of 2,588 with data, 4.8%). No clinical laboratory result required discontinuation of further chelation treatment in any child.

This is a retrospective analysis of clinical data with the associated limitations. No allowance for multiple testing was performed; false-positive results are possible. Analyses were nested by village of residence to account somewhat for unmeasured variables related to environmental exposure; however, there were undoubtedly other variables that may have influenced change in VBLL. The use of point-of-care testing equipment rather than ICPMS mildly decreased the accuracy of VBLLs. Our primary measure of DMSA effectiveness was ECP, which required an end-course VBLL within the defined timeframe, resulting in exclusion of five deaths during the first DMSA course, a possible source of bias. Another six deaths were in children with no courses meeting inclusion criteria. Exclusion of these deaths from 3,180 courses analysed is unlikely to have affected the primary outcome. To mitigate these exclusions, in our calculation of <2.5% mortality in severely poisoned children (initial VBLL ≥ 80 µg/dl) during the period studied we conservatively included all deaths considered caused or contributed to by lead poisoning (n = 11) in children who commenced any form of chelation.

In this retrospective analysis of an emergency intervention, oral DMSA as a single agent was a pharmacodynamically effective chelator for children with severe lead poisoning, even when administered to obtunded encephalopathic children by nasogastric tube. A lower end-course VBLL was associated with higher pre-course VBLL, a longer interval between chelation courses, and chelation administered in the inpatient setting. The reduction in VBLL achieved by courses of oral DMSA chelation that were entirely or predominantly inpatient was comparable to that observed in previously published clinical trials of oral DMSA or IV CaNa₂EDTA. Re-exposure to environmental lead contamination likely influenced the attenuated decline in VBLL associated with outpatient DMSA chelation. Decreased medication adherence is likely to have occurred in outpatient settings and may also have diminished the impact of chelation.

Although children with severe lead poisoning usually require multiple courses of DMSA chelation, the most desirable time interval between courses requires further investigation. This experience with basic supportive care and chelation in a large paediatric cohort adds significantly to the evidence base for clinical management of epidemic lead poisoning, particularly in resource-poor settings.