ABSTRACT
Following the endorsement of the Xpert MTB/RIF assay (Cepheid, Sunnyvale, CA, USA) by the World Health Organization (WHO) in 2010, Viet Nam’s National Tuberculosis Control Program (NTP) began using GeneXpert instruments in NTP laboratories. In 2013, Viet Nam’s NTP implemented an Xpert MTB/RIF external quality assurance (EQA) program in collaboration with the U.S. Centers for Disease Control and Prevention (CDC) and the Foundation for Innovative New Diagnostics (FIND). Proficiency-testing (PT) panels comprising five dried tube specimens (DTS) were sent to participating sites approximately twice a year from October 2013 to July 2016. The number of enrolled laboratories increased from 22 to 39 during the study period. Testing accuracy was assessed by comparing reported and expected results; percentage scores were assigned; and feedback reports were provided to sites. On-site evaluation (OSE) was conducted for underperforming laboratories. The results from the first five rounds demonstrate the positive impact of PT and targeted OSE visits on testing quality. On average, for every additional round of feedback, the odds of achieving PT scores of ≥80% increased 2.04-fold (95% confidence interval, 1.39- to 3.00-fold). Future work will include scaling up PT to all sites and maintaining the performance of participating laboratories while developing local panel production capacity.
INTRODUCTION
In 2015, 1.4 million people globally died of tuberculosis (TB), with >95% of deaths occurring in low- and middle-income countries (LMICs). Multidrug-resistant tuberculosis (MDR-TB) presents an increasing threat to TB control efforts (1). In 2015, according to World Health Organization (WHO) estimates, fewer than two-thirds of 10.4 million new TB cases, and just over a quarter of an estimated 480,000 new MDR-TB cases, were diagnosed and reported to the WHO (1). In Viet Nam, a lower-middle-income country (https://www.worldbank.org/en/country/vietnam) with high burdens of TB and MDR-TB (2), the fourth national antituberculosis drug resistance survey estimates that fewer than 20% of MDR-TB cases were diagnosed and enrolled for treatment, leading to intensified efforts by the National TB Control Program (NTP) to scale up case finding (3) so as to reach the WHO’s End TB Strategy targets for expanded access to rapid TB diagnostics and drug susceptibility testing for all TB patients (4). Viet Nam’s National Strategy for Tuberculosis Prevention and Control commits to proactively applying new technologies recommended by the WHO, in order to “universalize access to TB services including notification, treatments and prevention, as well as optimize the utilization of traditional methods in TB detection, diagnosis, treatment and prophylaxis so as to gain best results in Viet Nam conditions” (5).
In 2010, the WHO endorsed the Xpert MTB/RIF assay (Cepheid, Sunnyvale, CA, USA) for rapid detection of TB and rifampin (RIF) resistance, and this test is currently recommended as the initial diagnostic test for patients with suspected MDR-TB or HIV-associated TB and may be used as the initial test for all presumptive TB cases where resources allow (6). The Xpert MTB/RIF assay was rapidly incorporated into the national policies of many LMICs, and wide-scale implementation followed: since 2011, more than 16 million tests have been performed in 122 countries (7). In Viet Nam, by July 2016, the NTP had overseen the installation of 72 GeneXpert instruments (288 modules) at 59 sites, including 1 national, 5 regional, 47 provincial, and 6 district laboratories covering all provinces, with approximately 110,150 tests performed between January 2012 and July 2016 (personal communication).
Although the Xpert MTB/RIF assay has “game-changing” potential, implementation experience from resource-poor settings highlights the obstacles that limit impact, among them weaknesses in quality assurance (QA) activities, training and competency assessment, instrument verification and maintenance, supply chain management, service and technical support, connectivity and information technology (IT), and securing basic requirements (7–10). Yet the potential benefits of the Xpert MTB/RIF assay for patient management are most pronounced at peripheral levels of the health system and near the point of care, where quality management systems may be less developed (11–13). Like all diagnostic tests, the Xpert MTB/RIF assay should be supported by a quality assurance program, a key component of which is external quality assurance (EQA), incorporating on-site supervision and proficiency testing (PT) (14–16). PT is designed to assess all phases of diagnostic testing, from preanalytical to analytical and postanalytical processes. It involves periodically sending multiple well-characterized samples to members of a group of laboratories for analysis. Each laboratory’s results are compared with the expected results and those of other laboratories. Testing sites with discordant results are thus identified, and feedback is provided, including recommendations for corrective action and on-site supervisory visits conducted to address nonconformities. When accompanied by timely feedback and support with the performance of root cause analysis and implementation of corrective actions, PT has been shown to catalyze improvements in testing quality (16–20).
A range of options exist for Xpert MTB/RIF PT panels, including dried tube specimens (DTS) (U.S. Centers for Disease Control and Prevention [CDC]) developed using a methodology originally deployed in HIV proficiency-testing programs (18, 19), artificial sputum (WHO/Global Laboratory Initiative), dried culture spots (National Health Laboratory Service [NHLS], South Africa), and lyophilized and liquid samples. All panels have been found to be suitable for use in an Xpert MTB/RIF EQA program (21). To fulfill the requirements for PT, countries may choose to participate in a commercial PT scheme or may manufacture their own panels. The ability to prepare PT panels is largely dependent on technical capacity and the human resources available in the country. While commercial sources offer quality-assured panels and simplify procedures, the cost is high and may be prohibitive for many LMICs.
An Xpert MTB/RIF EQA program was implemented in Viet Nam from October 2013 to July 2016 by the CDC, the Viet Nam NTP, and the Foundation for Innovative New Diagnostics (FIND) as a key component of the laboratory quality improvement efforts, with the aim of evaluating the quality of testing being performed in the laboratory network (16). All the sites participating in the Xpert EQA program were at public sector health facilities, managed under the NTP. The sites included 1 national referral hospital, 4 regional referral hospitals (1 designated as a national hospital under the Ministry of Health but considered a regional referral facility by the NTP), 28 provincial hospitals, and 3 district hospitals. The annual testing volumes for the Xpert MTB/RIF assay at participating sites differed considerably, from a low of 33 tests (Thanh Xuan District, Hanoi, Viet Nam) to an upper limit of 10,913 tests (Pham Ngoc Thach Hospital, Ho Chi Minh City, Viet Nam) in 2016. This article describes the lessons learned during the implementation process and the impact of the program on the quality of testing among the first laboratories to use the Xpert MTB/RIF assay.
MATERIALS AND METHODS
PT panel production.Dried tube specimen (DTS) panels were created from dilutions of well-characterized Mycobacterium tuberculosis complex (MTBC) and nontuberculous mycobacterial (NTM) cultures grown in liquid culture medium (mycobacterial growth indicator tube [MGIT]; Becton, Dickinson, Sparks, MD), chemically inactivated with Xpert MTB/RIF sample reagent (SR) (Cepheid Inc., Sunnyvale, CA), and subjected to glass bead disruption. Cultures selected for inactivation ranged from 0.8 × 105 to 3.2 × 105 CFU/ml.
The inactivated culture was inoculated into MGIT medium and was incubated for 84 days at 37°C to verify inactivation. A 1:10 dilution of the inactivated culture was prepared, pipetted into 4-ml tubes, and dried at ambient temperature inside a biosafety cabinet. All procedures were conducted in a biosafety level III laboratory at the CDC in Atlanta, GA (15, 21) (Fig. 1). Each panel was composed of five unknown DTS samples. The CFU count per DTS was not determined. Amounts of detectable DNA differed from sample to sample, as evidenced by Xpert MTB/RIF cycle threshold results. The composition of each panel was unique; however, each panel included at least one of the following expected results: (i) TB not detected, (ii) TB detected, rifampin resistance not detected, and (iii) TB detected, rifampin resistance detected. Panels were packaged in plastic specimen transport bags, together with the standard operating procedure (SOP) for sample processing, result-reporting forms, and a letter with instructions on inspecting and testing samples and reporting results. These were sent as noninfectious shipments to the Viet Nam National TB Reference Laboratory (NRL) via international courier and were then sent by courier to participating laboratories with NTP-approved Vietnamese translations of each document. DTS samples were freshly prepared by the CDC for each round.
Procedure for manufacturing the Xpert MTB/RIF proficiency testing panel.
PT pilot: rounds 1 and 2.The first two rounds of PT constituted a pilot, which was conducted from October 2013 to November 2014. A total of 22 NTP testing sites were enrolled in round 1, and 23 were enrolled in round 2. All sites using the GeneXpert assay at the time were enrolled in the pilot. At laboratories, samples were rehydrated and tested according to the enclosed instructions. Results were manually entered into the reporting form, which was scanned and emailed to NRL coordinators, together with scanned copies of corresponding GeneXpert automated test reports. A manual cross-check of these documents was jointly performed by FIND and NTP coordinators in order to uncover any transcription errors.
Results and automated GeneXpert reports were compiled by the NRL in MS Excel and were emailed to the CDC for analysis and reporting. Scores were not assigned to pilot PT panels at the time but were retroactively applied for the purposes of analysis and comparison with rounds 3 to 5. Laboratory performance was assigned an accuracy score expressed as a percentage, with each of the five samples counting for 20 percentage points. Twenty points were allocated to correct determination of both MTB detection and RIF resistance compared with the expected result. Zero points were allocated to an incorrect determination of either MTB detection or RIF resistance, and 10 points were allocated to samples for which a RIF-indeterminate result was obtained. An unsuccessful result (error, invalid, or no result) received 5 points. Scores of 80% and above were considered satisfactory, and scores below 80% were considered unsatisfactory, thus allowing for one fully incorrect sample result. On-site evaluation (OSE) visits were jointly undertaken, by FIND and NTP assessors, to laboratories exhibiting any nonconformities in both rounds. Assessors used OSE visit checklists to identify the root causes of nonconformities and supported sites in the implementation of corrective actions.
A stakeholder meeting was held in April 2014. The NTP and FIND, with help from key TB partners in the country, presented on the program’s implementation requirements and the results of the pilot. The outcome was the NTP’s decision to proceed with scale-up of the EQA program as a key component in the strengthening of its TB laboratory network, with support from key partners.
PT program: rounds 3 to 5.As part of the ongoing Xpert MTB/RIF PT program, laboratories in Viet Nam routinely test one panel of five samples per site approximately twice a year. Three complete rounds took place between December 2014 and March 2016. Twenty-three laboratories were enrolled in round 3, and 39 were enrolled in rounds 4 and 5. Additional instruments were installed after the initiation of planning for PT rounds 4 and 5 and were therefore not included. Testing was performed and results reported as in pilot rounds. Scores were issued by the CDC in the form of individual feedback reports for each site, detailing the breakdown of scores by sample. These were returned by email to the NRL, where they were reformatted according to the NTP template, translated into Vietnamese, and sent by courier to testing sites.
The NRL and FIND jointly analyzed aggregate data and undertook OSE visits to laboratories with unsatisfactory scores (defined as a panel score of <80%), using OSE checklists to identify and address nonconformities as in pilot rounds. One additional OSE visit was made in round 3 to a laboratory with an 80% score in order to understand the need for OSE at sites meeting minimum requirements for successful completion of PT. No urgent technical nonconformities were identified. Additionally, general feedback was given following round 3 in the form of a 3-day workshop, attended by representatives from all sites and covering major nonconformities identified across foregoing rounds.
An unpaired t test was used to compare the mean proportions of laboratories returning a false-negative result when a majority of sites reported “very low” or “low” TB detection results. The percentage of false-negative results was then compared to the percentages of other unsuccessful results, including “errors,” “invalid,” and “no results.” We hypothesized that PT performance would improve with increasing experience with PT. We examined the association between PT performance (unsatisfactory, satisfactory, or perfect) and the number of times a lab had performed PT (including PT at the current round) using fixed- and mixed-effect proportional odds models (22) implemented in the ordinal R package (23). Although the laboratories represent sampling clusters, random effects were negligible, and therefore, we base inference on the fixed-effect model.
RESULTS
Five PT rounds, including the two pilot rounds, were completed between October 2013 and January 2016 as part of the Xpert EQA program in Viet Nam. The number of enrolled laboratories increased from 22 in round 1 and 23 in rounds 2 and 3 to 39 in rounds 4 and 5, representing all active testing sites at the time of the PT event. Overall, 93.8% (137/146) of PT panel results were returned to the CDC.
Of the 10 instances of nonparticipation, 1 was due to a shortage of cartridges, 2 to breakdown of the computer processing unit, and the remaining 7 to module failure. For the two sites with module failure, delays in repairs led to nonparticipation in two rounds of PT (rounds 4 and 5), with one site citing a lack of financial resources for module repair. All sites reporting module failure were diverting routine samples to alternative facilities included in the EQA activities.
The numbers of laboratories achieving satisfactory and unsatisfactory scores were determined for each round (Table 1). Round 1 saw the greatest proportion of unsatisfactory scores (18%) and the smallest proportion of perfect scores (36%). In subsequent rounds, the proportion of unsatisfactory scores remained within a tight range, between 5.9% and 9.1%, until round 5, during which all sites scored 100%.
Xpert MTB/RIF proficiency-testing scores from 36 laboratories in Viet Nam, 2013 to 2016
A comparison of the baseline and latest scores for each individual laboratory shows that sites either improved their testing accuracy or matched an original score of 100%. The median score improvement, from the first round in which a site participated to the latest round (round 5), was 10 percentage points for all sites, from 90% to 100%. Excluding sites that scored 100% at baseline, the median score increased by 20%, from an average of 80% to 90% at baseline. Scores did not increase uniformly in every round for every site, and there were only three instances of score decreases in subsequent rounds. All three occurred either between pilot rounds 1 and 2 or between rounds 2 and 3, when no sites had yet received either reports or feedback.
There were a total of 10 unsatisfactory results (scores of <80%) from nine laboratories, accounting for 7.3% of all available scores (Table 1). The majority (n = 6) belonged to sites participating for the first time. The remaining four cases involved sites that had previously taken part in a pilot round and so had not received feedback. Only one site obtained two unsatisfactory scores in a row; this site had a score of 40% during round 1, increasing to 70% in round 2. The instrument was subsequently removed from this laboratory and relocated to another laboratory, and therefore, it did not participate in subsequent rounds. Support in the form of OSE visits was provided to five underperforming sites as well as to one laboratory with a passing score of 80%. Where laboratories participated in a subsequent round, the median score increase immediately following an OSE visit was 30 percentage points, from 70% to 100%.
Issues identified during OSE visits to underperforming sites included overdue instrument calibration; inadequate monitoring and control of the temperatures of storage rooms, the laboratory, and refrigerators; improper use of the uninterruptible power source (UPS); and a lack of staff training. Poor understanding of NTP standard operating procedures (SOPs) is thought to have led to errors in sample preparation and in the addition of reagent to the cartridges. At one site, a new staff member had not been adequately trained. There were a number of technical issues with GeneXpert instruments, such as components returning an abnormal amplification curve, faulty fluid transfer, or loss of cartridge tube pressure. Assessors recommended reporting these errors to the manufacturer if they persisted. Other recommended corrective actions included scheduling instrument calibration; relocating instruments to a more appropriate space where the temperature could be controlled; improving supply chain management for the supply of cartridges; connecting or replacing the UPS; updating and maintaining patient log, storage temperature, workstation cleaning, and stock registers; and ensuring adherence to testing SOPs and GeneXpert quality indicators. In the case of one site, a higher-than-expected proportion of M. tuberculosis-negative results led assessors to recommend that laboratory managers review patient referral and enrollment criteria to ensure that test requests adhered to NTP guidelines. In the case of laboratories reporting module failure, the corrective action was completion and submission of an approval form for module repair.
Overall, 27 laboratories recorded at least one discordant result. There were a total of 51 discordant Xpert MTB/RIF test results, or 7.4% (51/685 available test results) (Fig. 2). Table 2 shows the frequency of each type of discordant result from underperforming sites (n = 21), as well as recommendations for addressing them provided during OSE visits to sites. False-negative M. tuberculosis results were the most common discordant finding. A positive association was found between the proportion of laboratories returning false-negative results and laboratories reporting “very low” semiquantitative M. tuberculosis detection results (P < 0.05). False-negative M. tuberculosis results may have been caused by small amounts of detectable M. tuberculosis DNA, which may have resulted from sample degradation caused by long shipment transit times, addition of insufficient sample volume to test cartridges, or other procedural errors, such as improper storage and testing temperatures of PT panels. The expected mean cycle threshold of probe A, or of probe C when that of probe A was 0, determined during the CDC’s panel validation, differed across all rounds. The mean cycle thresholds for rounds 1 to 5 ranged from 20.5 to 24.7, 19.3 to 23.5, 21.1 to 25.9, 19.3 to 22.7, and 19.7 to 21.6, respectively, all falling in the “medium” to “low” semiquantitative result range. The next-highest proportion of discordant results (2.5%) belonged to combined invalid, no-result, or error readings, followed by RIF-indeterminate samples (1.8% of M. tuberculosis-positive samples).
Types of discordant results obtained in the Viet Nam Xpert MTB/RIF proficiency-testing program, rounds 1 to 5.
Discordant results reported from sites with unsatisfactory proficiency-testing scores during the Viet Nam Xpert MTB/RIF proficiency-testing program and recommended corrective/preventive actions
The percentage of sites achieving perfect scores generally increased over the course of observation and reached 100% in our sample at the fifth round (Fig. 3). The odds of achieving scores of either ≥80% or 100% were estimated to be 2.04 (95% confidence interval [CI], 1.39 to 3.00) times greater with each new round of experience within sites. Our proportional odds regression model predicts that 93% of sites should be expected to achieve at least satisfactory scores (80%) after two rounds of prior experience with PT, and 91% should be expected to achieve perfect scores after four rounds of prior experience (Fig. 3).
Performance of laboratories in the Viet Nam Xpert MTB/RIF proficiency-testing program associated with prior rounds of testing experience. P(Score), proportion of sites achieving the specified score.
Turnaround times.Total round duration, from the time when the CDC sent panels to the time when the NRL forwarded feedback, was 250 days in round 3, decreasing to 119 days in round 4 and remaining relatively constant, at 123 days, in round 5. Meanwhile, across all rounds, the median laboratory turnaround time, from sites’ receipt of panels to reporting of results, was 6 days (interquartile range, 3 to 9.3 days). Nevertheless, only two laboratories exceeded the 30-day deadline, by 2 days and 1 day, in rounds 2 and 3, respectively.
The greatest delays occurred at the level of the NRL and the CDC. CDC processing times, from the receipt of results to the issuing of feedback reports, differed: the shortest processing time was 14 days, in round 4, and the longest was 119 days, in round 3. NRL processing times, from the receipt of feedback to the forwarding of feedback to sites, though high in round 3, at 80 days, has decreased across rounds, to 63 days in round 4 and 7 days in round 5. As a result, median feedback times, from the time when sites reported results to the time when they received feedback, were also reduced.
DISCUSSION
Program impact.The results from the first five rounds of the Xpert MTB/RIF EQA program in Viet Nam suggest that regular PT rounds, accompanied by feedback and tailored support to underperforming sites, contribute to laboratories’ improved overall testing accuracy and to reductions in the number of sites achieving unsatisfactory results. It is possible that experience, regardless of feedback, may also be important in improving the quality of testing. PT is a signal to labs that their performance is important and is being monitored, so personnel may become more diligent with each new experience with PT. Since round 5 culminated in all perfect scores, and continual effort is required to sustain laboratories’ performance, future rounds will provide insights into how testing accuracy can be maintained and what types of obstacles are encountered.
Quality assurance.The causes of nonconformities may be complex, and there is rarely a single fix. While DTS panels undergo robust validation prior to shipment to testing sites, the possibility still exists that some false-negative results may have been due to issues related to panel preparation. Feedback, OSE checklists, and assessors therefore approach issues holistically, attempting to cover all dimensions of Xpert MTB/RIF assay quality assurance. Except where technical assistance was required to address malfunctioning GeneXpert modules, corrective actions identified during OSE visits concerned laboratories’ interdependent quality management system activities: conducting staff training and refresher training in case of operator-related nonconformities, maintaining and updating registers, monitoring quality indicators, reviewing sample referral systems, and altering and updating facilities’ infrastructure. This underlines the need for a broader, ongoing QA program, of which the Xpert MTB/RIF EQA program is one component.
Timely feedback.Timely feedback is crucial to the program’s effectiveness and the continued commitment of sites. Delayed feedback may impede the timely implementation of essential corrective actions, prolonging poor testing accuracy and falsely informing the clinical management of patients. Feedback may lose its relevance, since sites are forced either to address issues in the interim or to delay the implementation of essential corrective actions. Challenges to rapid feedback cited in the literature focus on transportation and custom clearance issues, as well as the logistical complexity of sending panels and feedback to sites via intermediary NRLs, particularly where connectivity is an issue and paper-based communication is necessary (1). In Viet Nam, prompt feedback remains a challenge. In rounds 3 to 5, long turnaround times left little time for sites to process feedback and implement corrective actions before the next round of testing. In particular, some sites received round-4 panels less than a week after the receipt of round-3 feedback.
The much-longer feedback time in round 3 was caused by an NTP-wide shortage of cartridges, by the delay of reports by the NRL pending the development and approval of official NTP feedback forms translated into Vietnamese, and by a longer-than-average turnaround time at the CDC. After round 3, complete translated documentation was in place, and NRL feedback times were reduced dramatically across subsequent rounds. Meanwhile, CDC processing times remained variable and frequently long, owing to the scale of the multicountry program. Therefore, the introduction of an in-country PT program, including local manufacture of PT panels, is advocated to improve turnaround time, reduce costs, and build sustainability. To this end, the NTP has now developed in-country capacity for the manufacture of panels and has been managing the PT program since 2017.
Equipment breakdown.Equipment breakdowns were a challenge, as has been reported in other settings (7, 9). One site experienced machine breakdown in round 3, and another two in round 4. None had been repaired by round 5, so downtimes exceeded 6 to 15 months. Downtimes from module failures are common and cause significant disruptions to laboratories’ testing. This problem foregrounds the challenges of GeneXpert maintenance and servicing in LMICs, which include a lack of replacement modules in-country, the unforeseen financial burdens on countries for repairs, and low uptake of extended warranties due to lack of funding (7, 10). Albert et al. emphasize the need for strengthened networks of authorized service providers, including in peripheral settings (7). They also suggest the use of Web-based remote monitoring tools to assess instruments’ service and maintenance status, optimize inventory management, and facilitate remote support with installation and connectivity (7). Tracking would allow sites to anticipate problems and to take rapid action when they materialize. In the meantime, staff performing Xpert MTB/RIF testing should be trained to maintain and troubleshoot equipment using nationally approved training materials and, where problems are intractable, to contact the manufacturer for guidance and to seek formal approval for repairs from the relevant authority.
Program sustainability.NTP commitment is essential to the successful implementation and scale-up of Xpert EQA. In Viet Nam, there was some initial reluctance to devote scarce resources to Xpert EQA. However, the establishment of a technical working group and the convening of a stakeholder meeting to discuss the benefits and impact of the program were essential mechanisms to ensure that support for the program was effective; describing the nonconformities found at underperforming sites during pilot rounds was also important. A technical working group was established to enable rapid review of results and feedback forms; annual meetings were held to review PT results and nonconformities; and regional quality indicator monitoring was implemented. Additionally, the NTP is seeking ways of creating added incentives for satisfactory sites to implement corrective actions and achieve perfect scores. Discussions are under way to potentially revise NRL guidelines for Xpert EQA to include a higher success threshold. The transfer of technology for local manufacture of Xpert MTB/RIF PT panels is ongoing, and the scale-up of EQA to all GeneXpert testing sites was completed by the end of 2017 (personal communication). This work will be reported in a subsequent publication.
Conclusion.The results from the first five rounds of the EQA program show that there were improvements in overall scores across Xpert MTB/RIF EQA rounds, with the greatest improvements at previously underperforming sites that received supportive on-site visits. Further efforts should be directed toward improving and sustaining laboratories’ testing accuracy and reducing PT program turnaround and feedback times. Where nonconformities are identified during OSE visits, laboratories should be given detailed guidance on the implementation of sustainable quality management systems. With technical assistance from the CDC and FIND, the national TB program in Viet Nam has, since the time of this study, validated local production of PT panels at the National TB Reference Laboratory to establish a fully independent national EQA program so as to ensure sustainability, with the intention of expanding the EQA program to all testing sites as the program scales up nationally. Monitoring turnaround times and assessing further program impact as part of this transition will be important. With ongoing implementation of GeneXpert instruments in Viet Nam and worldwide, there is a corresponding need for expanded PT programs. The lessons learned in the implementation of the program in Viet Nam may prove helpful for LMICs seeking to replicate its success.
ACKNOWLEDGMENTS
We thank the laboratory managers and staff who participated in the program, and we acknowledge Tran Quynh Thi Hoang, CDC Vietnam, for her contribution to initiating the program. We are grateful to Yen Nguyen Thi Hoang for programmatic support and to Donatelle Erni for assistance with data analysis.
The findings and conclusions in this report are those of the authors and do not necessarily represent the official positions of the funding agencies.
This project has been supported by the President’s Emergency Plan for AIDS Relief (PEPFAR) through the Centers for Disease Control and Prevention (CDC) under the terms of cooperative agreement U2G PS002746.
FOOTNOTES
- Received 17 October 2018.
- Returned for modification 19 November 2018.
- Accepted 6 December 2018.
- Accepted manuscript posted online 19 December 2018.
This is a work of the U.S. Government and is not subject to copyright protection in the United States. Foreign copyrights may apply.
