The Bråten (AIM) 2019 trial was not a superior trial to Albert 2013: why antibiotic exposure changes the interpretation

Abstract

The 2019 Bråten AIM trial is often considered superior to Albert 2013 because it was newer, multicentre, and larger overall. This interpretation overlooks differences in population and antibiotic exposure. AIM included 118 patients with Modic change type 1 (MC1; 58 amoxicillin, 60 placebo), whereas Albert randomised 162 MC1 patients (90 co-amoxiclav, 72 placebo). AIM administered 0.75 g amoxicillin every eight hours; Albert evaluated 0.5 g and 1.0 g regimens. A new study-arm-level analysis identified a strong exposure-response association between oral dose, modelled intradiscal probability of target attainment (PTA), and 12-month improvements in pain and disability. Estimated MIC₉₀ PTA was 36.4% for AIM's regimen and 56.4% for 1.0 g every eight hours, meaning AIM achieved approximately 65% of the higher-dose PTA. AIM should therefore be interpreted as an intermediate-exposure trial, not a definitive test of higher-dose oral amoxicillin. Meta-analyses should account for dose and target-site exposure.

Albert vs Bråten

The 2019 AIM trial by Bråten and colleagues is often treated as the definitive test of antibiotics for chronic low back pain associated with Modic changes.¹ That interpretation usually rests on three observations: AIM was newer, multicentre, and recruited 180 participants.

From this perspective, AIM appears methodologically superior to the 2013 Albert trial,² and its conclusion - that amoxicillin did not provide a clinically important benefit - is often presented as outweighing the earlier positive result.

But this comparison overlooks two important questions:

1. Were the directly comparable patient populations actually larger in AIM?

2. Did the trials test equivalent antibiotic exposures?

The answer to both is NO.

A new study-arm-level exposure-response analysis provides a pharmacological explanation for the apparently conflicting outcomes (Czaplewski et al., 2026).³ Rather than showing that Albert's result failed to replicate, the evidence suggests that the trials occupy different positions on an antibiotic exposure-response curve.

AIM was not the larger study for the directly comparable population

AIM randomised 180 participants overall: 118 with Modic change type 1 (MC1) and 62 with Modic change type 2 (MC2). The groups were analysed both together and separately.

The Albert trial enrolled only patients with MC1. It randomised 162 participants: 90 to co-amoxiclav and 72 to placebo.

For the directly comparable MC1 population, therefore, the relevant sample sizes were Albert 162 vs Bråten 118. 

AIM was larger only when its biologically distinct MC1 and MC2 groups were combined.

For the MC1 population studied by Albert, the Albert trial was larger.

Being newer and multicentre remains relevant to study appraisal, but these characteristics do not automatically make AIM the more informative trial for estimating the effect of oral amoxicillin in MC1 patients.

The trials did not test the same antibiotic exposure

Albert evaluated co-amoxiclav at two amoxicillin doses:

• 0.5 g every eight hours

• 1.0 g every eight hours

The Albert 2013 publication pooled the two active-dose groups for its principal treatment comparison, although it reported a trend towards greater efficacy with the double dose. The new exposure-response analysis separated the dose groups using dose-specific data.

AIM used an intermediate regimen:

• 0.75 g amoxicillin every eight hours

AIM therefore did not reproduce Albert's highest-exposure regimen. It tested a dose between Albert's low- and high-dose groups and used treatment for 90 rather than 100 days.

This distinction is crucial. A trial of 0.75 g three times daily cannot determine the clinical result achievable with 1.0 g three times daily if intradiscal exposure increases meaningfully with dose.

Probability of target attainment reveals a substantial exposure gap

Probability of target attainment (PTA) estimates the proportion of patients expected to achieve a predefined pharmacodynamic target against Cutibacterium acnes at the proposed infection site.

See Czaplewski et al., 2026: Higher exposure to antibiotics is associated with greater clinical improvement of chronic low back pain with Modic changes

The exposure-response paper modelled free intradiscal amoxicillin exposure and estimated MIC₉₀ PTA as follows:

Oral regimen	Estimated MIC90 PTA
0.5 g every eight hours	12.95%
0.75 g every eight hours	36.42%
1.0 g every eight hours	56.35%

The AIM regimen achieved an estimated PTA of approximately 36.4%, compared with 56.4% for Albert's 1.0 g regimen. AIM's PTA was therefore only about 65% of the PTA achieved by the 1.0 g regimen.

Neither oral regimen reached the commonly used population benchmark of at least 90% PTA. This suggests that oral delivery itself may have difficulty producing reliable antibacterial exposure within a poorly vascularised disc.

Nevertheless, the difference between 36.4% and 56.4% is substantial. AIM tested a regimen expected to achieve the target in only a little over one-third of patients. It should therefore not be treated as an estimate of the maximum benefit potentially achievable with oral amoxicillin.

The missing variable was exposure

AIM reported a small treatment effect overall. At 12 months, the adjusted between-group difference in RMDQ was -1.6 points. In its MC1 subgroup, however, the difference was -2.3 points in favour of amoxicillin, with a 95% confidence interval of -4.2 to -0.4.

AIM regarded this as below its predefined threshold for clinical importance. Yet the direction and intermediate magnitude of the finding are consistent with what the exposure-response analysis would predict from an intermediate dose.

Findings of the new exposure-response study

Across oral study arms, higher nominal dose and higher modelled intradiscal PTA were both significantly associated with greater 12-month improvements in pain and disability. The associations were statistically significant in all four oral models.

Within the oral dataset:

• Nominal dose explained 99.9% of between-arm variation in pain improvement.

• Nominal dose explained 98.3% of between-arm variation in disability improvement.

• MIC₉₀ PTA explained 88.7% of variation in pain and 91.1% in disability.

This produces a coherent ordering:

placebo -> 0.5 g -> 0.75 g -> 1.0 g

Clinical improvement increased as estimated exposure increased. AIM did not sit outside this pattern. It occupied the intermediate exposure position and produced an intermediate clinical response.

That is not a failed replication of an equivalent intervention. It is a result consistent with a previously unrecognised exposure-response relationship.

The 1.0 g outcome was reproduced in a much larger cohort

The interpretation is strengthened by the 2017 open-label study, which treated 1,024 patients with co-amoxiclav at 1.0 g every eight hours for 100 days.⁴

Although an open-label cohort does not provide the same protection against bias as a randomised placebo-controlled trial, its outcome at the 1.0 g dose was highly consistent with the corresponding Albert 2013 dose group. The exposure-response pattern also remained directionally consistent when the large 2017 cohort was excluded from sensitivity analyses.

The 2017 study is not a substitute for randomisation. However, the reproducibility of the clinical change at the same dose is relevant evidence that should not be ignored when interpreting the oral literature.

What AIM can and cannot tell us

AIM remains an important, carefully conducted randomised trial. It provides valuable evidence about 750 mg amoxicillin three times daily in its selected population.

It does not, however, establish that:

• all oral amoxicillin regimens are clinically equivalent;

• 750 mg and 1,000 mg every eight hours produce equivalent intradiscal exposure;

• the 1.0 g regimen evaluated by Albert is ineffective;

• dose and target-site exposure can be disregarded when comparing trials.

A more precise conclusion is:

AIM found that 0.75 g amoxicillin every eight hours did not achieve its predefined clinically important benefit in the overall MC1/MC2 population. Its MC1 subgroup showed a smaller benefit than Albert 2013, consistent with its lower modelled intradiscal exposure.

This interpretation preserves AIM's findings without extending them beyond the regimen and population actually studied.

Why conventional meta-analyses can produce uncertain conclusions

Systematic reviews commonly pool antibiotic studies as though they were evaluating a single intervention.^5,6 But "antibiotics" is not an exposure level.

Combining trials that used different doses, schedules, treatment durations, drugs, routes, and patient populations can obscure a genuine exposure-response relationship. Averaging suboptimal regimens with higher-exposure regimens may underestimate the outcome associated with the latter.

The oral evidence base includes regimens with estimated PTA values far below the conventional 90% benchmark.

Meta-analyses that do not account for this may primarily answer:

What is the average outcome across several incompletely exposed regimens?

That is different from asking:

What outcome is associated with a regimen that achieves adequate antibacterial exposure in the disc?

The uncertain conclusions of previous reviews may therefore reflect pharmacological heterogeneity rather than irreconcilable clinical evidence.

A better hierarchy for interpreting the evidence

For questions about the effect of the higher oral amoxicillin dose in MC1 patients, Albert 2013 is the more directly relevant randomised study:

• It included 162 MC1 participants, compared with 118 in AIM's MC1 subgroup.

• It evaluated the 1.0 g every-eight-hours regimen.

• It identified a trend towards greater efficacy at the higher dose.

• Its 1.0 g outcome was reproduced in the large 2017 open-label cohort.

AIM should be interpreted as the principal randomised study of the intermediate 0.75 g regimen, not as a higher-quality replacement for every dose examined by Albert.

Important limitations

The exposure-response findings are compelling but not definitive proof of efficacy. The analysis used study arms rather than individual-patient data, included a small number of studies, and relied on modelled rather than directly measured intradiscal exposure. Shared study characteristics and other differences between trials could contribute to the observed pattern.

The findings should therefore be considered hypothesis-strengthening evidence. They support a pharmacologically coherent reinterpretation of the literature and identify dose and intradiscal target attainment as variables that future trials must address directly.

This analysis also does not justify routine or unsupervised antibiotic treatment for chronic low back pain. Prolonged antibiotic use carries individual risks and contributes to antimicrobial resistance. Clinical use requires appropriate evidence, patient selection, safety oversight, and antimicrobial stewardship.

The conclusion should change

The oral antibiotic literature should no longer be reduced to "Albert was positive, whereas the superior AIM trial was negative."

The more accurate interpretation is:

• Albert 2013 was the larger randomised study in the directly comparable MC1 population.

• AIM used a lower dose than Albert's 1.0 g regimen and achieved only about 65% of its modelled PTA.

• AIM's intermediate clinical outcome was consistent with its intermediate exposure.

• The 1.0 g outcome was reproducible in the 2017 open-label cohort.

• Differences in pain and disability outcomes were strongly associated with dose and modelled intradiscal exposure.

AIM remains valuable evidence, but it is not an adequate estimate of what can be achieved with the higher oral regimen. For oral amoxicillin in MC1 disease, Albert 2013 should remain the reference randomised study for the 1.0 g three-times-daily dose.

The next generation of studies should not ask merely whether an antibiotic was administered. They should establish whether the regimen can achieve an appropriate pharmacodynamic target within the disc.

1          Bråten LCH, Rolfsen MP, Espeland A, et al. Efficacy of antibiotic treatment in patients with chronic low back pain and Modic changes (the AIM study): double blind, randomised, placebo controlled, multicentre trial. BMJ 2019; 367: l5654.

2          Albert HB, Sorensen JS, Christensen BS, Manniche C. Antibiotic treatment in patients with chronic low back pain and vertebral bone edema (Modic type 1 changes): a double-blind randomized clinical controlled trial of efficacy. Eur Spine J 2013; 22: 697–707.

3          Czaplewski L, Gilligan C, McHale D. Higher exposure to antibiotics is associated with greater clinical improvement of chronic low back pain with Modic changes. 2026; published online Sept 1. DOI:https://doi.org/10.21203/rs.3.rs-10748663/v1.

4          Albert HB. Antibiotic treatment of patients with chronic low back pain and Modic changes. Schmerzmed 2017; 33: 49–52.

5          Wong AYL, Mallow GM, Pinto SM, et al. The efficacy and safety of oral antibiotic treatment in patients with chronic low back pain and Modic changes: A systematic review and meta-analysis. JOR Spine 2024; 7: e1281.

6 Liu C, Abdel Shaheed C, Braten LC, et al. Antibiotic treatment for low back pain, radicular pain, or both. Cochrane Database Syst Rev 2026; 4: CD014221.

FAQs

Was the AIM trial larger than Albert 2013?

AIM was larger only when patients with Modic change types 1 and 2 were combined. In the directly comparable Modic type 1 population, AIM included 118 participants, whereas Albert 2013 included 162.

Did AIM replicate Albert's highest amoxicillin dose?

No. AIM used 0.75 g every eight hours. Albert evaluated 0.5 g and 1.0 g every eight hours.

What was the estimated PTA for each regimen?

The estimated MIC₉₀ PTA was 36.4% for 0.75 g every eight hours and 56.4% for 1.0 g every eight hours. AIM's PTA was therefore approximately 65% of that of the higher-dose regimen.

Does the exposure-response analysis prove antibiotics are effective?

No. It is a study-arm-level analysis based partly on modelled exposure and cannot establish causality. It provides a pharmacologically coherent explanation for heterogeneity and supports further adequately exposed trials.

Why might previous meta-analyses be inconclusive?

Pooling regimens with substantially different doses and target-site exposure can dilute the apparent outcome of higher-exposure treatment. Future evidence syntheses should account for dose, schedule, route, and PTA.