GeneXpert Omni Implementation Trials – Assessing Impact and Minimizing Bias

Adithya Cattamanchi, MD, MAS

GeneXpert Omni

• Single-cartridge, point-of-care platform

• Low power consumption (solid-state)

• Integrated battery (4 hours) + supplemental battery (12 hours)

• Automatic connectivity

Rapid, onsite molecular testing at primary care clinics in high-burden countries

Trials of GeneXpert impact

• Five primary care clinics in S. Africa, Tanzania, Zambia, and Zimbabwe • Compared POC Xpert testing vs. sputum smear microscopy

• 20 clusters (lab + 2 primary care clinics) in S. Africa • Compared Xpert testing vs. sputum smear microscopy

GeneXpert trials – summary of key findings

• Bacteriologically-confirmed cases increased

• Time to treatment of confirmed TB decreased

• Patient outcomes largely unchanged – TB NEAT

• No difference in morbidity score in TB culture-positive patients who had begun treatment

• No difference in all-cause mortality (8% in both groups) • Decreased pre-treatment loss to follow-up (8% vs. 15%)

– XTEND • No difference in all-cause mortality (3.9% vs. 5%) • No difference in pre-treatment loss to follow-up (17% vs. 14.9%)

Why no impact?

• High rates of empiric treatment – TB NEAT: 22% (17% in Xpert arm; 26% in microscopy arm)

– XTEND: about 10% in both arms

• Offsets gains in sensitivity and impact associated with Xpert

• Trial design: How to account for empiric treatment?

Menzies NA et al. Lancet ID 2014

Design choices – bias towards the null?

• TB NEAT and XTEND are among the best examples of diagnostic intervention trials

• Designing any trial involves a series of tradeoffs

• Key questions for next round of trials

– Could specific design choices have biased results toward the null?

– Can we give greater weight to avoiding bias towards the null when making design choices?

Design choices: bias towards null? TB NEAT

Design choice Potential impact

Individual patient randomized trial •Higher pre-test probability of TB lower threshold for empiric treatment

Patients screened, consented and enrolled by study staff

•Reduce loss to follow-up

Offered VCT for HIV and CXR while awaiting test results

• Increase likelihood of treatment at initial visit • Increase in empiric treatment rates

TB culture performed on all patients • Increase in bacteriologic confirmation in smear arm > Xpert arm •Minimize delay in treatment initiation in

smear arm > Xpert arm

Active follow-up at 2 months and 6 months

•Minimize loss to follow-up • Increase opportunity for empiric treatment

Primary outcome (morbidity) assessed among culture-positive patients starting treatment

•Decrease morbidity in smear arm > Xpert arm

Design choices: bias towards null? XTEND

Design choice Potential impact

Patients screened, consented and enrolled by study staff

•Reduce loss to follow-up

CXR, sputum culture for HIV-positives if index test negative

• Increase in empiric treatment rates • Increase in bacteriologic confirmation in

smear arm > Xpert arm •Minimize delay in treatment initiation in

smear arm > Xpert arm

Phone follow-up at 1 week and 1, 2 and 4 months; air-time vouchers/home visits

•Minimize loss to follow-up

Clusters randomized within geographic strata only

•Other co-variates unbalanced lower power

Other issues to consider - 1

• Power for assessing mortality as the primary outcome

– Mortality is uncommon in primary care settings

– Prohibitively large sample size required to detect small but meaningful effect sizes

Possible solutions: Combined endpoints, meta-analysis

TB NEAT

XTEND

Cox et al

I-V Overall (I-squared = 0.0%, p = 0.941)

study

TBNEAT(1)

Xtend(1)

D+L Overall

Cox et al.

0.88 (0.68, 1.14)

ES (95% CI)

0.94 (0.59, 1.50)

0.85 (0.56, 1.27)

0.88 (0.68, 1.14)

0.87 (0.54, 1.40)

100.00

%

(I-V)

30.66

40.18

29.16

Weight

0.88 (0.68, 1.14)

ES (95% CI)

0.94 (0.59, 1.50)

0.85 (0.56, 1.27)

0.88 (0.68, 1.14)

0.87 (0.54, 1.40)

100.00

%

(I-V)

30.66

40.18

29.16

Weight

Xpert reduces risk of mortality Xpert increases risk of mortality

1.5 2

I-V Overall (I-squared = 0.0%, p = 0.941)

study

TBNEAT(1)

Xtend(1)

D+L Overall

Cox et al.

0.88 (0.68, 1.14)

ES (95% CI)

0.94 (0.59, 1.50)

0.85 (0.56, 1.27)

0.88 (0.68, 1.14)

0.87 (0.54, 1.40)

100.00

%

(I-V)

30.66

40.18

29.16

Weight

0.88 (0.68, 1.14)

ES (95% CI)

0.94 (0.59, 1.50)

0.85 (0.56, 1.27)

0.88 (0.68, 1.14)

0.87 (0.54, 1.40)

100.00

%

(I-V)

30.66

40.18

29.16

Weight

Xpert reduces risk of mortality Xpert increases risk of mortality

1.5 2

0.94 (0.59, 1.50)

0.85 (0.56, 1.27)

0.87 (0.54, 1.40)

0.88 (0.68, 1.14) Overall

Favors Xpert Favors no Xpert

Peter JG et al. Effect on mortality of point-of-care, urine-based lipoarabinomannan testing to

guide tuberculosis treatment initiation in HIV-positive hospital inpatients: a pragmatic, parallel-

group, multicountry, open-label, randomised controlled trial. Lancet 2016.

Khaki A et al. The effect of automated nucleic acid amplification assays on mortality in routine

care settings: meta-analysis of individual participant data. Union World Lung Health

Conference, 2015.

IPD meta-analysis of Xpert RCTs RR 0.88 (0.68, 1.14) [3 RCTs, 8140 participants, 399 deaths]

LAM RCT RR 0.82 (0.70, 0.96) [1 RCT, 2528 participants, 578 deaths]

Other issues to consider - 2

• How pragmatic should a pragmatic trial be?

– Omni has potential to improve health center processes contributing to losses to follow-up

– Differences between arms minimized if study-related processes impact routine processes

StaRI Draft Checklist:

(StaRI: Standards for Reporting of Implementation studies)

• Patients should be recruited to the service and not to the study

• Outcomes should be assessed using routinely collected data

Pinnock H et al. Imp Sci 2015

XPEL TB Trial Design Xpert Omni Performance Evaluation to facilitate Linkage to care

Adithya Cattamanchi Margaret Handley

Sara Ackerman

Achilles Katamba Moses Joloba

David Moore Katherine Fielding

David Dowdy Luke Davis

Steering Committee: Gerald Friedland (Yale), Andrew Ramsay (WHO/TDR), Madhukar Pai (McGill), Noah Kiwanuka (MakCHS), Grant Theron (Stellenbosch) Funding: NIH/NHLBI R01 HL130192; FIND/Cepheid RFA for devices/cartridges

Frank Mugabe

Specific Aims

• Aim 1: To compare patient outcomes at health centers randomized to GeneXpert Omni vs. standard-of-care TB diagnostic evaluation strategies.

• Aim 2: To identify processes and contextual factors that influence the effectiveness of GeneXpert Omni.

• Aim 3: To compare the costs and epidemiological impact of GeneXpert Omni vs. standard-of-care TB diagnostic evaluation strategies

Comparison of intervention and control arms

VIS

IT 1

Standard-of-Care

(CONTROL ARM)

All Patients Patients with HIV

Onsite Omni

(INTERVENTION ARM)

All Patients

Treat if Xpert positive*

Perform Xpert testing

Collect sputum

Treat if Xpert positive*

Collect sputum

Refer for Xpert testing

Collect sputum

Prepare/examine smear

Collect sputum

Prepare/examine smear

Smear-positive: Treat

Treat if Xpert positive*

Smear-negative: Collect sputum and

refer for Xpert testing

VIS

IT 2

V

ISIT

3

VIS

IT 4

* Empiric treatment or additional testing at clinician’s discretion

Design Overview

• Cluster-randomized trial

• Sites (N=20-24): Primary care clinics Inclusion Criteria – Perform sputum smear microscopy – Participate in NTP-sponsored EQA – Send samples to a central site for Xpert testing Exclusion Criteria – Do not agree to be randomized – Perform smear examination on <150 patients per year – Diagnose <15 smear-positive TB cases per year

• Participants (N≈8000): All adults initiating TB evaluation over 2-year period Exclusion Criteria – Have sputum collected as part of active, community-based case finding – Extra-pulmonary TB only

Outcomes

• Primary – Proportion treated for microbiologically-confirmed TB

within 2 weeks

• Secondary – Number tested for TB – Number and proportion with confirmed TB – Number and proportion treated for TB (overall and if

confirmed) – Mortality + pre-treatment LTFU at 6 months – Incremental cost-effectiveness ratio – 10-year TB incidence

Procedures

• Patient-level data using routine data sources (Aim 1) – TB registers, ART register and Xpert referral forms

• Photos uploaded to secure server monthly

– GxAlert server download

• Patient vital status (Aim 1) – Phone call or home visit 6 months after initial health center visit – Intensive tracing of participants unable to be contacted

• Focused data collection during quarterly site visits (Aim 2) – Patient surveys to assess costs and satisfaction with care (N=20-40/site) – Provider surveys to assess attitudes, self-efficacy

• Health system costing studies (Aim 3)

-6 0 24 30 Pre-trial Trial Complete

follow-up

Randomization

• Stratification – reduce between cluster variation – Health centers randomized within strata defined by

primary outcome

• Restriction – balance between arms of site- and patient-level covariates – HIV prevalence among patients evaluated for TB

– Health center size (volume of presumed TB patients)

– Health center region (four quadrants of Uganda)

– Proportion treated empirically

Summary

• Lack of impact of GeneXpert to date is in part due to the technology

• Rapid, onsite molecular testing could improve factors that contribute to missed and delayed diagnosis/treatment

• Omni implementation trials should be as pragmatic as possible

Questions/Comments?