1. The Impact of Expanding ART-Eligibility on Overall Treatment Initiation and Retention in Care: An Application of Regression Discontinuity in Zambia
Aaloke Mody, MD
Division of HIV, ID, and Global Medicine
University of California, San Francisco
September 28, 2018
UCSF CFAR
Closing the Gap between Rigor and Relevance: Methodological Opportunities for Implementation Science

2. Motivation

3. Motivation “Treat all” is being adopted globally based on RCTs
HPTN 052 – ART reduces transmission
INSIGHT START and TEMPRANO – Improved patient outcomes with immediate versus deferred treatment
RCTs establish the biological effects of ART  Efficacy
Real world effects of adopting this policy remain unknown  Effectiveness
Ulimate public health impact also depends on the behavior of individual patients and capacity of health systems to absorb new patients
Benefits of expanding eligibility may be
Direct  ART initiation
Indirect  Improved retention
Negative spillover effects could include increased congestion and crowding out of sicker patients
WHO HIV Treatment Guidelines 2015; Over 2017; Abdool Karim NEJM 2015; Fox PLOS Med 2017

4. Quasi-Experimental Designs: Rigor and Relevance
Quasi-experimental designs can approximate experimental conditions in real world data
Regression Discontinuity
Instrumental Variable
Interrupted Time Series
Difference in Differences
Fixed Effect
Relevance
Rigor
Observational
RCT
Quasi-Experiment
Study interventions delivered through real-life systems in every-day contexts
Assess effectiveness (i.e., the effect of actually expanding ART versus the biological effect of ART)
Evidence with high rigor AND relevance in the right setting
Rigor depends on ability to meet the underlying assumptions of the study design
Relevance depends on being able to answer the right question

5. Motivation
Zambia previously changed its HIV treatment guidelines* on April 1, 2014 to expand ART eligibility to include:
Patients with a CD4 between 350 and 500 cells/L
All pregnant and breastfeeding women under Option B+
We leverage this 2014 guideline change in a quasi-experimental study design to assess the effects of this expansion change and shed light onto what might be expected when expanding to “treat all”
Quasi-experimental study designs offer an underused strategy to address these challenges by approximating experimental conditions in real world data
An example of this is regression discontinuity which has been previously used in South Africa to assess the effects of ART eligibility in those with a CD4 below 350
Click
Zambia adopted universal treatment in December 2016, but prior to this, it changed its HIV treatment guidelines on April 1, 2014 to expand ART eligibility to include patients with a CD4 between 350 and 500 as well as pregnant and breastfeeding women and their partners under Option B+.
We leverage this 2014 change in Zambia’s HIV treatment guidelines in a quasi-experimental study design to shed light onto what might be expected when expanding to “treat all”
*Adopted treat-all in December 2016
Zambia HIV Guidelines 2010, 2014

6. Study Objectives
To estimate the real world effects of expanding treatment eligibility on ART initiation and retention in care in the overall clinic population using regression discontinuity design.
To quantify the effect of initiating ART in routine care on retention using an instrumental variable design.
Hypothesis: Expanding ART-eligibility will lead to improvements in both ART initiation and retention in care without crowding out sicker patients

7. Methods

8. Study Population Measurements
ART naïve patients greater than 15 years old
Newly enrolling before and after the change in Zambia’s treatment guidelines (August 1, 2013 to November 1, 2014)
64 clinics supported by the Centre for Infectious Disease Research in Zambia (CIDRZ) in Lusaka, Eastern, Western, and Southern Provinces
Measurements
EMR system used in routine HIV care in Zambia (SmartCare)

9. Regression Discontinuity Design (RDD)
RDD can be used when treatment is assigned by an arbitrary threshold
Compare patients just above and below that threshold (local)
Since cutoff is arbitrary, assume patients are similar on measured and unmeasured characteristics, thus, exposure is as if random
Bor Epidemiology 2014, Barnighausen J Clin Epi 2017

10. Regression Discontinuity Design (RDD)
RDD to estimate the effects of implementing new ART guidelines
Compare patients enrolling just before and after the guideline change
Outcomes
ART Initiation by 3 months
Retention in Care at 6 months
Visit between 3 and 9 month
In care and on ART at 6 months
Composite Outcome
Model
Estimated risk difference at the time of implementation using a modified Poisson regression with robust variances
Excluded patients enrolling:
90 days prior to guideline rollout to avoid bias from “cross-over”
60 days after to account for a gradual guideline rollout
Sensitivity analyses for choices in model specification (results not shown)
Bor Epidemiology 2014

11. RD Instrumental Variable (IV) Analysis
We estimated the effect of actually initiating ART on retention in care using implementation of new guidelines as an instrumental variable
In observational settings, many measured and unmeasured common causes (i.e., confounders) of ART initiation and retention in care
IVs allow for unbiased estimates even with unmeasured confounding between the treatment (ART initiation) and outcome (retention) under certain assumptions:
IV is associated with the treatment
There are no common causes of the IV and the outcome
The IV only affects the outcome through the treatment
Swanson Epidemiology 2013

12. Patient Subgroups* Always Eligible: Eligible by 2010 Guidelines
CD4 less than or equal to 350 cells/L
WHO Stage 3 or 4
Active Tuberculosis
Newly Eligible: Eligible by 2014 Guidelines, but not 2010 Guidelines
CD4 between 350 and 500 cells/L
Pregnant or Breastfeeding Women
Not Yet Eligible: Not Eligible by 2010 or 2014 Guidelines
CD4 greater than 500 cells/L
*Defined at enrollment, but irrespective of the actual date of enrollment

13. Results

15. Patient Characteristics at Enrollment, n=34,857
Gender
Male Sex, n (%)
13,635 (39.1%)
Non-pregnant Female
13,998 (40.2%)
Pregnant or Breastfeeding Female
7,224 (20.7%)
Median Age, years (IQR)
34 (28, 41)
Median CD4 count, cells/L (IQR)
268 (134, 430)
WHO Stage, n (%) I
16,437 (55.5%) II
5,804 (19.6%)
III
6,748 (22.8%) IV
604 (2.0%)
Tuberculosis in past 6 months, n (%)
1,587 (4.6%)
Eligibility Subgroup, n (%)
Always Eligible
20,819 (70.1%)
Newly Eligible
5,578 (18.8%)
Not Yet Eligible
3,319 (11.2%)
Province, n (%)
Lusaka
18,532 (53.2%)
Eastern
6,201 (17.8%)
Southern
4,893 (14.0%)
Western
5,231 (15.0%)

16. PATIENT CHARACTERISTICS AT ENROLLMENT, n=34,857
Always Eligible
Newly Eligible
Not Yet Eligible
Pre-Guidelines
n=7,159
Post-Guidelines n=6,528
n=1,726
Post-Guidelines n=1,899
n=997
Post-Guidelines n=1,107
Gender, n (%)
Male
3164 (44.2)
2923 (44.8)
408 (23.6)
466 (24.5)
344 (34.5)
378 (34.1)
Female
2652 (37.0)
2499 (38.3)
615 (35.6)
627 (33.0)
653 (65.5)
729 (65.9)
Pregnant or Breastfeeding Female
1343 (18.8)
1106 (16.9)
703 (40.7)
806 (42.4)
Median Age, years (IQR) 35
(29, 41)
(30, 42) 31
(25, 38)
(26, 38) 32
(27, 39) 33
(27, 40)
Median CD4 count, cells/L (IQR)
183
(93, 275)
181
(92, 272)
436
(389, 486)
440
(393, 491)
628
(558, 760)
637
(560, 772)
WHO Stage, n (%) 1
2775 (43.0)
2473 (42.8)
1120 (80.5)
1229 (79.1)
665 (78.5)
747 (81.2) 2
1099 (17.0)
1052 (18.2)
272 (19.5)
325 (20.9)
182 (21.5)
173 (18.8) 3
2343 (36.3)
2069 (35.8)
0 (0) 4
234 (3.6)
179 (3.1)
Recent TB, n (%)
570 (8.0)
534 (8.2)

17. Mody PLOS Medicine 2018

18. All Patients: ART Initiation by 3 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
ART Initiation, % (95% CI)
56.4
(54.5, 58.4)
70.1
(68.4, 71.8)
+13.6
(11.1, 16.2)
<0.001
Mody PLOS Medicine 2018

19. All Patients: Retention in Care at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
Retention in Care, % (95% CI)
64.9
(63.0, 66.8)
69.1
(67.3, 70.8)
+4.1
(1.6, 6.7)
0.001
Mody PLOS Medicine 2018

20. All Patients: In Care and on ART at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
In Care on ART, % (95% CI)
46.9
(45.0, 48.9)
57.7
(55.9, 59.6)
+10.8
(8.1, 13.5)
<0.001
Mody PLOS Medicine 2018

21. Always Eligible: ART Initiation by 3 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
ART Initiation, % (95% CI)
72.4
(70.2, 74.6)
78.5
(76.5, 80.6)
+6.2
(3.2, 9.2)
<0.001
Mody PLOS Medicine 2018

22. Always Eligible: Retention in Care at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
Retention in Care, % (95% CI)
72.8
(70.5, 75.0)
73.2
(71.0, 75.4)
+0.4
(-2.7, 3.6)
0.790
Mody PLOS Medicine 2018

23. Always Eligible: In Care and on ART at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
In Care on ART, % (95% CI)
61.0
(58.6, 63.4)
65.6
(63.3, 68.0)
+4.6
(1.2, 8.0)
0.008
Mody PLOS Medicine 2018

24. Newly Eligible: ART Initiation by 3 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
ART Initiation, % (95% CI)
31.8
(27.1, 36.5)
75.5
(71.4, 79.5)
+43.7
(37.5, 49.9)
<0.001
Mody PLOS Medicine 2018

25. Newly Eligible: Retention in Care at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
Retention in Care, % (95% CI)
62.0
(57.0, 66.9)
75.6
(71.7, 79.6)
+13.6
(7.3, 20.0)
<0.001
Mody PLOS Medicine 2018

26. Newly Eligible: In Care and on ART at 6 months
Always Eligible
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
In Care on ART, % (95% CI)
27.1
(22.5, 31.7)
62.6
(58.2, 67.0)
+35.5
(29.2, 41.9)
<0.001
Mody PLOS Medicine 2018

27. Not Yet Eligible: ART Initiation by 3 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
ART Initiation, % (95% CI)
20.2
(14.5, 25.9)
32.7
(27.0, 38.5)
+12.5
(4.5, 20.6)
0.0023
Mody PLOS Medicine 2018

28. Not Yet Eligible: Retention in Care at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
Retention in Care, % (95% CI)
52.9
(46.2, 59.5)
57.1
(51.2, 63.0)
+4.3
(-4.6, 13.2)
0.349
Mody PLOS Medicine 2018

29. Not Yet Eligible: In Care and on ART at 6 months
Pre-Guidelines
Post-Guidelines
Risk Difference
p-value
In Care on ART, % (95% CI)
16.6
(11.5, 21.6)
27.4
(22.0, 32.8)
+10.8
(3.4, 18.3)
0.0042
Mody PLOS Medicine 2018

30. Instrumental Variable Analysis: Effect of ART Initiation on Retention in Care
Actually initiating ART led to a 37.9% increase in retention in care (95% CI 28.8, 46.9%, p<0.001)
2.6 patients need to be initiated on ART to prevent one episode of LTFU (95% CI 2.1, 3.5)
Mody PLOS Medicine 2018

31. Limitations Results only generalizable to current care setting
May not apply to linkage or initiation under different circumstances
Unable to assess viral suppression
Retention is an imperfect proxy for virologic outcomes
Excluded patients enrolling immediately before or after the guideline change
Results were robust to model specification in sensitivity analyses and no evidence of short-term secular trends
Limitations of primary data source
Potential for misclassification of subgroup/outcome

32. Conclusion

33. Conclusions Effects of expanding ART eligibility
All Patients: 13.6% increase in ART initiation, 4.1% increase in retention, and 10.8% increase in those in care and on ART at 6 months
Effects primarily driven by improvements in ART initiation and retention in Newly Eligible patients
Small increases in ART initiation and the percentage in care and on ART in Always Eligible and Not Yet Eligible patients.
Potentially due to overall expansion of ART supply and an informal loosening of eligibility criteria for patients not yet officially eligible
Initiating ART in routine care increased retention by 37.9%
70% of patients still present at a more advanced stage
25% of eligible patients are not initiated on ART by 3 months
40% are not in care and on ART at 6 months

34. Implications
Adopting “treat all” will likely lead to similar improvements in the number of patients on ART and retained in care
There was no evidence of increased clinic congestion or crowding out of sicker patients
Efforts still needed to diagnose and link patients to care earlier and then keep them retained in care
Overall effects of expanding ART eligibility alone may be modest
Further evidence still needed on real world effects of “treat all”
Overall impact on viral suppression
Better understanding of those that never start ART or quickly drop out
Targeted interventions for those that don’t benefit from eligibility alone
Quasi-experimental designs are a powerful tool for understanding the effects of real world implementation of evidence-based interventions

35. Thank you! UCSF Elvin Geng Diane Havlir Monica Gandhi Monika Roy
Nancy Czaicki (in memorium)
UC Berkeley
Nancy Padian
Georgetown University
Charles Holmes
CIDRZ
Izukanji Sikazwe
Carolyn Bolton-Moore
Kombatende Sikombe
Thea Savory
Mwanza wa Mwanza
Jake Pry
Anjali Sharma
Arianna Zanolini
Paul Somwe
Zambian Ministry of Health
Bill and Melinda Gates Foundation

36. Results of Regression Discontinuity Sensitivity Analyses
Final Model
100 Day Bandwidth
50 day Bandwidth
30 Day Transition Period
Adjusted
Weighted
Risk Difference
95% CI
All Patients
ART Initiation
13.6
11.1 – 16.2
13.5
10.4 – 16.7
16.1
11.4 – 20.7
11.1
8.6 – 13.5
14.6
12.0 – 17.3
10.8 – 16.4
Retention in Care
4.1
1.6 – 6.7
5.4
2.2 – 8.5
4.9
0.3 – 9.6
3.4
0.9 – 5.8
3.8
1.1 – 6.5
4.5
1.7 – 7.3
In Care on ART
10.8
8.1 – 13.5
11.2
7.9 – 14.5
12.3
7.5 – 17.1
8.9
6.3 – 11.4
12.0
9.1 – 14.9
10.6
7.7 – 13.6
IV Estimate
37.9
28.8 – 46.9
31.3
20.0 – 42.7
38.3
21.4 – 55.2
38.6
29.5 – 47.7
35.5
26.4 – 44.7
35.1
24.7 – 45.5
Always Eligible
6.2
3.2 – 9.2
6.4
2.7 – 10.1
8.3
2.9 – 13.7
2.0 – 7.7
5.9
2.9 – 8.9
6.0
2.7 – 9.3
0.4
-2.7 – 3.6
1.1
-2.7 – 5.0
0.3
-5.4 – 5.9
-0.3
-3.3 – 2.7
-2.7 – 3.5
0.1
-3.3 – 3.5
4.6
1.2 – 8.0
4.7
0.5 – 8.9
6.5
0.4 – 12.7
3.7
0.5 – 6.9
1.1 – 7.9
0.4 – 7.8
Newly Eligible
43.7
37.5 – 49.9
39.4
31.7 – 47.0
40.0
28.5 – 51.6
34.1 – 46.0
46.4
40.2 – 52.6
40.9
34.1 – 47.7
7.3 – 20.0
14.5
6.7 – 22.2
11.0
-0.4 – 22.3
12.8
6.7 – 18.8
14.4
8.1 – 20.7
13.8
6.9 – 20.7
29.2 – 41.9
33.3
25.5 – 41.1
31.0
19.9 – 42.1
32.1
26.1 – 38.1
38.4
32.1 – 44.7
26.4 – 40.2
Not Yet Eligible
12.5
4.5 – 20.6
15.9
6.5 – 25.4
16.6
3.2 – 30.0
7.0
-0.3 – 14.2
9.2
1.4 – 17.0
5.9 – 23.0
4.3
-4.6 – 13.2
10.4
-0.3 – 21.1
-1.8 – 29.0
-8.4 – 8.5
2.5
-6.4 – 11.4
7.7
-1.9 – 17.3
3.4 – 18.3
15.0
6.6 – 23.5
14.3
2.6 – 26.0
5.6
-1.0 – 12.2
1.0 – 15.7
12.7
5.0 – 20.4