Evaluating the impact of an Xpert® MTB/RIF- based TB diagnostic algorithm in a routine operational setting in Cape Town

Date
2017-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH SUMMARY: Decades of reliance on slow, inaccurate diagnostic tests have contributed to poor case detection and impeded tuberculosis (TB) control efforts globally. The development of an accurate, rapid molecular diagnostic test, Xpert® MTB/RIF (Cepheid, Sunnyvale, CA, USA) (Xpert), offers the prospect of identifying more cases, detecting them rapidly and enabling quicker treatment initiation. Xpert is a nucleic acid amplification test that simultaneously detects genetic sequences for Mycobacterium tuberculosis complex and the presence of mutations conferring resistance to rifampicin. Xpert sensitivity is substantially higher than smear microscopy (88% compared to 53.8% for a single smear) and provides a test result within a day (compared to 8-16 days for liquid culture). Whilst laboratory and demonstration studies suggest that Xpert has the technical capacity to address the limitations of conventional smear and culture tests, very little is known about how this translates into patient and public health benefits in routine operational conditions. The overall aim of this thesis was to undertake rigorous scientific research into the impact of an Xpert® MTB/RIF-based TB diagnostic algorithm in a routine operational setting in Cape Town. This entailed a pragmatic comparison between the existing smear/culture-based TB diagnostic algorithm and the newly introduced Xpert-based algorithm. The magnitude and range of benefits for laboratory confirmed cases of TB and MDR-TB were assessed. Impact analysis was guided by the Impact Assessment Framework which ensured a systematic and comprehensive approach to the evaluation of the new diagnostic algorithm. This framework addresses five aspects of impact: Effectiveness Analysis assesses the impact on the numbers of cases diagnosed and appropriately started on treatment as well as the timeliness of results and of treatment initiation. Equity Analysis assesses whether marginalised groups who may be more affected benefit from the new test – poor people, women and HIV-infected specifically. Health Systems Analysis assesses the human resource, laboratory infrastructure, procurement and quality assurance implications. Scale-up Analysis assesses the economic costs and benefits of scaling up the new technology from both a provider and a patient perspective. Horizon Scanning assesses what other similar technologies are available or likely to become available and how these compare in their projected performance. The stepped-wedge analysis of TB yield (Chapter 2) in five sub-districts between 2010 and 2013 showed that among the 54,393 presumptive cases tested, the proportion with a bacteriological diagnosis of TB was not increased in the Xpert-based algorithm. We found a decline in TB yield over time, possibly attributable to a declining TB prevalence. When the time-effect was taken into consideration, there was no difference TB yield – yield was 19.3% (95% CI 17.7% to 20.9%) in the Xpert-based algorithm compared to 19.1% (95% CI 17.6% to 20.5%) in the smear/culture-based algorithm with a risk difference of 0.3% (95% CI -1.8% to 2.3%, p=0.796). Inconsistent implementation of the Xpert-based algorithm and the frequent use of culture tests in the smear/culture-based algorithm may have contributed to the yield parity. The multidrug-resistant (MDR)-TB yield study (Chapter 3) found that amongst the 10,284 TB cases identified in the five sub-districts, the Xpert-based algorithm was more effective in identifying MDR-TB than the smear/culture-based algorithm. Pre-treatment, there was a higher probability of having drug susceptibility tests undertaken (RR=1.82, p<0.001) and of being diagnosed with MDR-TB (RR=1.42, p<0.001) in the Xpert-based algorithm than in the smear/culture-based algorithm. Overall 8.5% of TB cases were detected with MDR-TB in the Xpert-based algorithm compared to 6% in the smear/culture-based algorithm, translating to approximately 375 additional MDR-TB cases diagnosed in Cape Town annually. The study on TB treatment initiation and treatment success undertaken in five sub-districts in October – December 2011 (Chapter 4) found that a higher proportion of cases initiated TB treatment in the Xpert group (84%, 508/603) than in the smear/culture group (71%, 493/693, p<0.001). The adjusted odds ratio for treatment initiation in the Xpert group was 1.98 (p<0.001). Cases >44 years in age (AOR=0.49, p<0.001) and previously treated cases (AOR=0.64, p=0.020) were less likely to initiate treatment. Laboratory delay was associated with non-initiation (AOR=0.96 per day, p<0.001). The reduction in TB treatment delay from a median of 15 days in the smear/culture group to 7 days in the Xpert group did not translate into improved TB treatment outcomes and treatment success rates were 80% in both groups (AOR=0.95, p=0.764). The MDR-TB treatment commencement study (Chapter 5) undertaken in 10 high TB burden facilities found that the time from test taken to treatment initiation was reduced from 43 days in the smear/culture-based algorithm (n=375) to 17 days in the Xpert-based algorithm (n=120) with a mean reduction of 25 days (p<0.001). Median laboratory turnaround time from test taken to result available in the laboratory was reduced from 24 days to <1 day with a mean reduction of 20 days (P<0.001) between algorithms. The qualitative study on MDR-TB patient pathways (Chapter 6) showed that patients experienced substantial delays before being diagnosed – these delays may not have been reflected using the data from the laboratory and clinics. Avoidable health system delays resulted from providers not testing for TB at initial health contact, non-adherence to testing algorithms, results not being available and failure to promptly recall patients with positive results. Negative perceptions of the public sector (as over-burdened, with long waiting times, negative staff attitudes and lack of privacy) were prevalent and contributed to deferred health-seeking, interruptions to the diagnostic process and to patient’s preferential use of the private sector, contributing to delays in both algorithms. The MDR-TB patient costing study (Chapter 7) assessed direct (out-of-pocket expenses) and indirect costs (lost productivity costs for patient’s time) incurred. The median patient cost from initial health visit to treatment initiation was reduced from $68.1 in the smear/culture-based algorithm to $38.3 (p=0.004) in the Xpert-based algorithm. Median direct costs were low at $6.7 and $4.4 (p=0.321) respectively. The difference in costs was attributable to time costs as the median number of visits to MDR-TB treatment was reduced from 20 in the smear/culture-based algorithm to 7 in the Xpert-based algorithm (p<0.001). Further details are provided below in the section on equity. From a laboratory costing perspective (Chapter 8) we found a 43% increase in overall PTB laboratory costs at the central laboratory, from $440,967 in the smear/culture-based algorithm to $632,262 in the Xpert-based algorithm for 3-month periods. The cost per TB case diagnosed increased by 157% from $48.77 in the smear/culture-based algorithm to $125.32 in the Xpert-based algorithm. The mean total cost per MDR-TB case diagnosed was similar at $190.14 in the smear/culture-based algorithm compared to $183.86 in the Xpert-based algorithm. From an effectiveness perspective, the Xpert-based algorithm did not result in an increase in the number TB cases diagnosed or improve treatment outcomes amongst those initiating treatment. It did however significantly reduce treatment delay and increased the proportion of TB cases initiating treatment. The Xpert-based algorithm resulted in a higher proportion of MDR-TB cases being diagnosed and reduced MDR-TB treatment commencement time. From an equity perspective the Xpert-based algorithm helped reduce health inequities through improving effectiveness as described above. However, these benefits did not shield patients from economic losses. The proportion unemployed increased (from symptom onset to the time of the interview) in both groups: from 39% to 73% in the smear/culture group (p<0.001) and from 53% to 89% in the Xpert group (p<0.001). From symptom onset to the time of the interview there was a 16% decrease in median household income in the smear/culture group and 13% decrease in the Xpert group and “catastrophic” costs were experienced by 38% and 27% (p=0.165) in respective groups who lost >10% of monthly household income. Health system failures at several levels from poor initial planning for Xpert implementation to human resource and IT infrastructure deficits, to poor accountability and inefficient service delivery as well as low community preparedness are likely to have diminished the full potential impact of the Xpert-based algorithm. Urgent attention needs to be paid to these issues to optimise the benefit of Xpert. From a scale-up perspective the increase in laboratory costs in our study are offset to some extent by the cost-saving to MDR-TB patients. As part of broader work we have developed a discrete event simulation model and validated it using the results from the studies presented in this thesis. This model will be used to evaluate more cost-effective diagnostic options and the benefits of a more sensitive test such as Xpert Ultra, which our horizon scanning suggests is the most likely current replacement for Xpert. These studies have limitations. It was difficult to control for bias - for example the non-random allocation of facilities to different study arms was ouside our control. Generalisability to other settings, especially rural settings, is limited as these studies were undertaken within a well-resourced, urban setting, with relatively good health and laboratory infrastructure. It was possible to address temporal trends in some studies (for example the stepped-wedge analysis of TB yield) but not in others (for example the MDR-TB treatment commencement study where decentralization of services may have contributed to the findings). The studies presented in this thesis have several novel aspects: they were undertaken at the level of the Xpert-based diagnostic algorithm and not the individual test, reflecting how tests were used in clinical practice. They reflect the patient, provider and health system factors that influenced outcomes and that are essential to understanding the impact of the new diagnostic algorithm in routine programmatic conditions. In addition, the use of Impact Assessment Framework provided a comprehensive view of the benefits and limitations of Xpert. These studies highlight the effect of the early introduction of new tools into under-prepared and inefficient health systems and provide insights into some of the health system weaknesses that could be addressed to optimise the impact of Xpert. Unless concerted efforts are made to address these weaknesses, the investment in this expensive new technology will not provide the full range of benefits possible.
AFRIKAANSE OPSOMMING: Omdat daar oor dekades heen op stadige, onakkurate diagnostiese toetse staatgemaak is, het die opsporing en beheer van tuberkulose (TB) wêreldwyd daaronder gely. Die ontwikkeling van ’n akkurate, vinnige molekulêre diagnostiese toets, Xpert® MTB/RIF (Cepheid, Sunnyvale, CA, VSA) (Xpert), bring die moontlikheid dat meer gevalle nou dalk vinniger geïdentifiseer kan word sodat behandeling gouer kan begin. Xpert maak van nukleïensuurversterking gebruik om terselfdertyd genetiese reekse vir Mycobacterium tuberculosis kompleks sowel as weerstandigheidsmutasies teen rifampisien op te spoor. Xpert is meer sensitief as smeermikroskopie (88% vergeleke met 53.8% vir ’n enkele smeer), en die toetsresultaat is binne ’n dag beskikbaar (vergeleke met 8-16 dae vir vloeistofkweking). Hoewel laboratorium- en demonstrasiestudies daarop dui dat Xpert oor die tegniese vermoë beskik om die beperkings van konvensionele smeer- en kwekingstoetse te bowe te kom, is weinig nog bekend oor die werklike voordele wat dit in normale bedryfsomstandighede vir pasiënte en openbare gesondheid inhou. Die oorkoepelende doel met hierdie tesis was om streng wetenskaplike navorsing te onderneem oor die impak van ’n Xpert® MTB/RIF-gebaseerde algoritme vir TB-diagnose in ’n normale bedryfsomgewing in Kaapstad. Hiervoor is ’n pragmatiese vergelyking onderneem van die bestaande smeer/kwekingsgebaseerde algoritme vir TB-diagnose, en die nuut ingestelde Xpert-gebaseerde algoritme. Die omvang van én verskeidenheid voordele vir laboratoriumbevestigde TB- en MDR-TB-gevalle is beoordeel. Impakontleding is deur die impakbeoordelingsraamwerk gerig, wat ’n stelselmatige en omvattende benadering tot die beoordeling van die nuwe diagnostiese algoritme verseker het. Hierdie raamwerk ondersoek vyf aspekte van impak: Doeltreffendheidsontleding beoordeel die impak op die getal gevalle wat gediagnoseer word en met gepaste behandeling begin, sowel as die tydigheid van resultate en behandelingsaanvang. Billikheidsontleding beoordeel of gemarginaliseerde groepe wat dalk erger geraak word – in die besonder arm mense, vroue en MIV-geïnfekteerde persone – by die nuwe toets baat vind. Gesondheidstelselontleding beoordeel die implikasies vir menslike hulpbronne, laboratoriuminfrastruktuur, verkryging en gehalteversekering. Opskaleringsontleding beoordeel die ekonomiese koste en voordele verbonde aan die opskalering van die nuwe tegnologie uit sowel ’n verskaffer- as ’n pasiënteoogpunt. Horisonbespieding beoordeel watter ander soortgelyke tegnologieë beskikbaar is of waarskynlik beskikbaar sal kom, en hoe die verwagte prestasie daarvan van Xpert s’n verskil. Die trapsgewyse wigontleding van TB-opbrengs (hoofstuk 2) in vyf subdistrikte tussen 2010 en 2013 toon dat onder die 54 393 vermoedelike gevalle wat getoets is, die persentasie met ’n bakteriologiese TB-diagnose nie met die Xpert-gebaseerde algoritme verhoog het nie. Die navorsing dui op ’n afname in TB-opbrengs oor tyd, moontlik as gevolg van ’n afname in TB-voorkoms. Toe die tydeffek in ag geneem is, was daar geen verskil in TB-opbrengs nie – 19.3% (95% CI 17.7% tot 20.9%) met die Xpert-gebaseerde algoritme vergeleke met 19.1% (95% CI 17.6% tot 20.5%) met die smeer-/kwekingsgebaseerde algoritme, met ’n risikoverskil van 0.3% (95% CI -1.8% tot 2.3%, p=0.796). Inkonsekwente implementering van die Xpert-gebaseerde algoritme en die gereelde gebruik van kwekingstoetse in die smeer-/kwekingsgebaseerde algoritme kon tot die pariteit in opbrengs bygedra het. Die studie van multimiddelweerstandige (MDR-) TB-opbrengs (hoofstuk 3) bevind dat onder die 10 284 TB-gevalle wat in die vyf subdistrikte geïdentifiseer is, die Xpert-gebaseerde algoritme MDR-TB doeltreffender as die smeer-/kwekingsgebaseerde algoritme gediagnoseer het. Voor behandeling, was die waarskynlikheid dat middelweerstandigheidstoetse gedoen sal word (RR=1.82, p<0.001) en dat MDR-TB gediagnoseer sal word (RR=1.42, p<0.001), hoër met die Xpert-gebaseerde algoritme as met die smeer-/kwekingsgebaseerde algoritme. Die Xpert-gebaseerde algoritme het 8,5% van TB-gevalle as MDR-TB geïdentifiseer, vergeleke met 6% wat deur die smeer-/kwekingsgebaseerde algoritme geïdentifiseer is. Dít kom neer op die diagnose van sowat 375 bykomende MDR-TB-gevalle in Kaapstad per jaar. Die studie van TB-behandelingsaanvang en -behandelingsukses wat van Oktober tot Desember 2011 in vyf subdistrikte onderneem is (hoofstuk 4), het bevind dat ’n hoër persentasie in die Xpert-groep met TB-behandeling begin het (84%, 508/603) as in die smeer-/kwekingsgroep (71%, 493/693, p<0.001). Die waarskynlikheid van behandelingsaanvang was hoër in die Xpert-groep (AOR=1.98, p<0.001). Gevalle bo 44-jarige ouderdom (AOR=0.49, p<0.001) en voorheen behandelde gevalle (AOR=0.64, p=0.020) het ’n laer waarskynlikheid getoon om met behandeling te begin. Laboratoriumvertraging het ’n verband met die gebrek aan behandelingsaanvang getoon (AOR=0.96 per dag, p<0.001). Die daling van ‘n mediaan van 15 dae in TB-behandelingsvertraging in die smeer-/kwekingsgroep tot 7 dae in die Xpert-groep het nie in die praktyk tot beter TB-behandelingsuitkomste gelei nie, en behandelingsuksessyfers was 80% in albei groepe (AOR=0.95, p=0.764). Die studie van MDR-TB-behandelingaanvangtyd (hoofstuk 5) wat in 10 fasiliteite met ’n swaar TB-las onderneem is, bevind dat die tydsduur vandat die toets gedoen word totdat behandeling begin, verkort is van 43 dae met die smeer-/kwekingsgebaseerde algoritme (n=375) tot 17 dae met die Xpert-gebaseerde algoritme (n=120), met ’n gemiddelde verkorting van 25 dae (p<0.001). Die mediane laboratoriumomkeertyd vandat die toets geneem is totdat die resultaat beskikbaar was in die laboratorium, is verkort van 24 dae tot <1 dag, met ’n gemiddelde verkorting van 20 dae (p<0.001) tussen algoritmes. Die kwalitatiewe studie van MDR-TB-pasiëntbehandelingsroetes (hoofstuk 6) toon dat pasiënte beduidende vertragings ervaar voordat hulle gediagnoseer word – hierdie vertragings kom moontlik nie na vore uit die data van die laboratorium en klinieke nie. Voorkombare gesondheidstelselvertragings kan daaraan toegeskryf word dat verskaffers nie met die eerste kontakbesoek reeds vir TB toets nie, dat toetsalgoritmes nie nagekom word nie, dat resultate nie beskikbaar is nie, en dat verskaffers versuim om pasiënte met positiewe resultate dadelik te laat terugkeer. Negatiewe opvattings oor die openbare sektor (soos oorlading, lang wagtye, negatiewe personeelingesteldheid en ’n gebrek aan privaatheid) is algemeen en het bygedra tot die uitstel van die soeke na gesondheidshulp, onderbrekings in die diagnostiese proses, en pasiënte se voorkeur vir die privaat sektor, wat tot vertragings in albei algoritmes gelei het. Die studie van MDR-TB-pasiëntkoste (hoofstuk 7) het direkte koste (uitgawes uit die pasiënt se sak) sowel as indirekte koste (die pasiënt se tydkoste vir verlore produktiwiteit) beoordeel. Die mediane pasiëntkoste van die eerste gesondheidsbesoek tot en met behandelingsaanvang is verminder van $68.1 met die smeer-/ kwekingsgebaseerde algoritme tot $38.3 (p=0.004) met die Xpert-gebaseerde algoritme. Die mediane direkte koste was laag teen $6.7 en $4.4 (p=0.321) onderskeidelik. Die verskil in koste kan toegeskryf word aan tydkoste aangesien die mediane getal besoeke tot en met MDR-TB-behandeling verminder is van 20 met die smeer-/kwekingsalgoritme tot 7 met die Xpert-gebaseerde algoritme (p<0.001). Die ekonomiese impak op pasiënte word hieronder in die afdeling oor billikheid bespreek. Uit die oogpunt van laboratoriumkoste (hoofstuk 8) dui die studie op ’n toename van 43% in algehele PTB-laboratoriumkoste by die sentrale laboratorium, van $440,967 met die smeer-/kwekingsgebaseerde algoritme tot $632,262 met die Xpert-gebaseerde algoritme oor tydperke van 3 maande. Die koste per gediagnoseerde TB-geval het met 157% toegeneem van $48.77 met die smeer-/kwekingsgebaseerde algoritme tot $125.32 met die Xpert-gebaseerde algoritme. Die gemiddelde totale koste per gediagnoseerde MDR-TB-geval was soortgelyk, naamlik $190.14 met die smeer-/kwekingsgebaseerde algoritme vergeleke met $183.86 met die Xpert-gebaseerde algoritme. Wat doeltreffendheid betref, het die Xpert-gebaseerde algoritme nie tot ’n toename in die getal gediagnoseerde TB-gevalle óf beter behandelingsuitkomste onder diegene wat met behandeling begin het, gelei nie. Dit het egter behandelingsvertraging beduidend verkort en die persentasie TB-gevalle wat met behandeling begin het, verhoog. Die Xpert-gebaseerde algoritme het daartoe gelei dat ’n groter persentasie MDR-TB-gevalle gediagnoseer is, en het MDR-TB-behandelingaanvangtyd verkort. Wat billikheid betref, het die Xpert-gebaseerde algoritme gesondheidsonbillikheid help verminder deur doeltreffendheid te verbeter, soos wat hierbo beskryf is. Tog het hierdie voordele nie pasiënte teen ekonomiese verliese beskerm nie. Die persentasie werklose persone in albei groepe het toegeneem (van aanvang van simptome tot en met tyd van die onderhoud): van 39% tot 73% in die smeer-/kwekingsgroep (p <0.001) en van 53% tot 89% in die Xpert-groep (p <0.001). Van die aanvang van simptome tot en met die tyd van die onderhoud was daar ’n afname van 16% in die mediane huishoudelike inkomste in die smeer-/ kwekingsgroep, en ’n afname van 13% in die Xpert-groep. Altesaam 38% en 27% (p=0.165) in die onderskeie groepe het “katastrofiese” koste ondervind en het sodoende meer as 10% van hulle maandelikse huishoudelike inkomste verloor. Mislukking van gesondheidstelsels op verskeie vlakke, van swak aanvanklike beplanning vir Xpert-implementering, en tekorte in menslike hulpbronne en IT-infrastruktuur, tot swak verantwoordbaarheid, ondoeltreffende dienslewering en swak gemeenskapsgereedheid, het waarskynlik gekeer dat die Xpert-gebaseerde algoritme sy volle potensiële impak gehad het. Hierdie kwessies verg dringende aandag om die voordele van Xpert te optimaliseer. Wat opskalering betref, word die toename in laboratoriumkoste in hierdie studie in ’n sekere mate geneutraliseer deur die kostebesparing vir MDR-TB-pasiënte. As deel van ’n groter projek is ’n diskrete gebeurtenissimulasiemodel ontwikkel en met behulp van die resultate van die studies in hierdie tesis bekragtig. Hierdie model sal gebruik word vir die beoordeling van meer kostedoeltreffende diagnostiese moontlikhede, sowel as van die voordele van ’n gevoeliger toets soos Xpert Ultra, wat volgens die horisonbespieding tans die mees waarskynlike plaasvervanger vir Xpert blyk te wees. Hierdie studies het bepaalde beperkings. Dit was moeilik om vir sydigheid te kontroleer – die nie-lukrake toewysing van fasiliteite aan verskillende afdelings van die studie was byvoorbeeld buite die navorsers se beheer. Veralgemeenbaarheid na ander omgewings, veral landelike omgewings, is beperk omdat hierdie studies in ’n stedelike omgewing met goeie hulpbronne en betreklik goeie gesondheids- en laboratoriuminfrastruktuur onderneem is. Tydtendense kon in party studies in ag geneem word (byvoorbeeld die trapsgewyse wigontleding van TB-opbrengs), maar nie in ander nie (byvoorbeeld die studie van MDR-TB-behandelingaanvangstyd, waar desentralisasie van dienste moontlik tot die bevindinge bygedra het). Die studies in hierdie tesis bevat verskeie nuwe en oorspronklike aspekte: studies is op die vlak van die Xpert-gebaseerde diagnostiese algoritme in plaas van die individuele toets onderneem, en weerspieël hoe toetse in kliniese praktyk gebruik word. Dit reflekteer die pasiënt-, verskaffer- en gesondheidstelselfaktore wat uitkomste beïnvloed en noodsaaklik is om die impak van die nuwe diagnostiese algoritme in normale programmatiese omstandighede te verstaan. Daarbenewens bied die gebruik van die impakbeoordelingsraamwerk ’n omvattende blik op die voordele en beperkings van Xpert. Hierdie studies beklemtoon die effek van die vroeë bekendstelling van nuwe toetse in swak toegeruste en ondoeltreffende gesondheidstelsels, en bied insig in van die swakpunte in gesondheidstelsels wat aangespreek behoort te word om die impak van Xpert te optimaliseer. Tensy doelbewuste pogings aangewend word om hierdie swakpunte te verbeter, sal die belegging in hierdie duur nuwe tegnologie nie die volle omvang van moontlike voordele oplewer nie.
Description
Thesis (PhD)--Stellenbosch University, 2017.
Keywords
Xpert® MTB/RIF- based -- Impact analysis, Rifampicin -- Resistance -- Testing, Mycobacterium tuberculosis -- Molecular diagnosis, Diagnostic algorithm, Health care system -- Management -- Cape Town, UCTD
Citation