For a person newly diagnosed with both HIV and active tuberculosis, one of the first operational questions a treating clinician faces has nothing to do with which drugs to prescribe — it is when to start them. Beginning antiretroviral therapy (ART) too early can trigger a dangerous inflammatory reaction against the mycobacterium; starting it too late leaves a severely immunosuppressed patient exposed for weeks. A run of large randomized trials in the last two decades settled the question with more precision than most global-health guidance ever gets, and the answer turns out to hinge almost entirely on one number: the CD4 count at diagnosis.
This piece works through what that evidence actually showed, where the trials disagreed, and why the World Health Organization eventually simplified its own recommendation rather than keep asking field programs to run a CD4-based decision tree under outbreak conditions.
The CD4 Threshold That Sets the Clock
Guidance built on the trial evidence below converges on a small number of timing bands. For patients with a CD4 count under 50 cells/mm³, ART should start as soon as possible, ideally within the first two weeks of beginning TB treatment — the mortality risk in this group is high enough that the benefit of early viral suppression outweighs the added risk of an inflammatory reaction. For patients with a CD4 count of 50 cells/mm³ or higher, the recommended window widens to two to eight weeks; several of the studies below found no survival advantage to starting inside two weeks in this less-immunosuppressed group, and no measurable harm from waiting slightly longer either. Pregnant patients are advised to start as early as feasible regardless of CD4 count, both for their own treatment and to reduce transmission risk to the infant. Patients with drug-resistant TB are advised to start ART within the first eight weeks of TB treatment regardless of CD4 count, reflecting the longer and more complex treatment course that drug-resistant disease requires.
By March 2021, WHO guidance had moved toward a further simplification: starting ART within two weeks of TB treatment initiation regardless of CD4 count, in part to reduce the logistical burden on programs that would otherwise need same-day CD4 testing capacity to route a patient correctly.
What the Trials Actually Showed
Three trials did most of the work behind this guidance, and they did not all agree with each other in the same population.
| Trial | Population (CD4) | Arms compared | Key mortality/morbidity finding |
|---|---|---|---|
| SAPIT | <500 cells/mm³ | ART within 4 weeks of TB treatment start vs. within 4 weeks of completing the intensive phase vs. within 4 weeks after finishing TB treatment | 56% reduction in mortality with ART during TB treatment vs. after; mortality rate rose from 5.4 to 12.1 per 100 person-years when ART was delayed |
| STRIDE (ACTG A5221) | <250 cells/mm³ | ART within 2 weeks vs. 8–12 weeks of TB treatment start | No significant overall difference in AIDS/death; but in the CD4<50 subgroup, immediate ART cut AIDS-or-death events to 15.5% vs. 26.6% in the delayed arm |
| CAMELIA | ≤200 cells/mm³ (median 25) | ART at 2 weeks vs. 8 weeks of TB treatment start | 34% reduction in risk of death (18% vs. 27%) with earlier ART, in a population with very advanced immunosuppression |
Read together, the pattern is consistent even though the trials used different CD4 cutoffs and comparison windows: the mortality benefit of early ART concentrates almost entirely in patients with the lowest CD4 counts. STRIDE’s overall result — no significant benefit across its full <250 population — is often the headline finding cited, but its own pre-specified subgroup analysis at CD4<50 lines up closely with what SAPIT and CAMELIA found in similarly immunosuppressed cohorts.
The IRIS Trade-Off Programs Accept
None of this comes free. TB-associated immune reconstitution inflammatory syndrome (IRIS) — a paradoxical worsening of TB symptoms as a recovering immune system reacts against residual mycobacterial antigen — occurred more often, and sooner, in every early-ART arm across all three trials:
- SAPIT recorded IRIS in 20.1% of patients on the early-integrated schedule versus 8.4% on the sequential schedule.
- STRIDE recorded TB IRIS in 10.4% of the immediate arm versus 4.7% of the later arm, rising to 11.5% in the CD4<50 subgroup; IRIS also appeared sooner with earlier ART (a median of 29 days versus 82 days), though no IRIS-associated deaths occurred in the study.
- CAMELIA found the risk of TB IRIS significantly elevated in the early arm (hazard ratio 2.51).
The consistent finding across all three trials is that this increase in IRIS incidence did not erase the survival benefit of early ART, even though IRIS itself can require hospitalization or corticosteroid treatment in more severe presentations, particularly when the central nervous system is involved. That is the trade-off underlying the whole guidance structure: accept a higher, earlier, and generally manageable rate of inflammatory reaction in exchange for a substantially lower risk of death in the weeks that matter most.
Why a Single Rule Beats a CD4-Based Branch in the Field
A CD4-stratified protocol reads cleanly on paper: test, branch, treat accordingly. In a field program without same-day CD4 access, or where a second visit cannot be guaranteed, that branch becomes a delay, and delay is the variable the trial evidence most consistently punishes in the sickest patients. Collapsing the guidance toward a single two-week starting point removes the decision point most likely to fail under real operating conditions, at the cost of accepting a marginally higher average IRIS rate across the whole treated population — a trade a program can absorb far more easily than a missed CD4 threshold.
Conclusion
The evidence on when to start ART during TB treatment is unusually well resolved for a global-health question: start early, especially at CD4 counts under 50 cells/mm³, accept a higher and largely manageable rate of TB IRIS, and treat late initiation as the higher-risk default rather than the cautious one. The simplification toward a uniform two-week window reflects less a change in the underlying biology than a recognition that field programs perform better on one rule they can follow reliably than on a more finely tuned one they cannot always apply in time.