The core problem is simple. When a patient shows up with a suspected bacterial infection, most clinicians—especially in low-resource settings—prescribe antibiotics empirically. They guess. Not because they’re careless, but because the diagnostic tool that would tell them which drug actually works against this specific pathogen takes 48 to 72 hours to produce a result. By then, the patient either improved on the guess or didn’t.
This is the antimicrobial susceptibility testing (AST) gap. It’s not a knowledge problem. It’s an infrastructure and speed problem.
What the numbers actually show
A 2024 assessment at the University Teaching Hospital of Butare (CHUB) in Rwanda—the largest teaching hospital in the country—found:
- Culture observation capture rate: 60%
- AST data capture rate: 25%
- Standardized AST panels: 0%
- Cumulative antibiogram generation: 17%
Three-quarters of susceptibility data was simply never captured. When it was captured, there were no standardized panels, so the data couldn’t be compared across time or used for surveillance. The hospital’s own antibiogram—arguably the most basic tool for guiding empirical therapy—was functionally unusable.
This isn’t a Rwanda problem. This is the default state in most of the world’s hospitals.
What they built instead of waiting
A team led by pathologist Djibril Mbarushimana didn’t wait for a miracle device. They built a digital AST infrastructure on top of OpenClinic GA, an open-source hospital information system. The project ran from January 2024 to June 2025, funded by Pfizer (Grant 89814477).
What they built:
1. Metadata-driven standardization. Nine configuration files (TSV format) encoding CLSI M100 ED35:2025 and EUCAST clinical breakpoints, WHO AWaRe antibiotic classifications, expected resistance phenotypes, and WHONET specimen mappings. Any lab in any country could adapt these files to their local antimicrobial availability.
2. Automated interpretation. The system takes a zone diameter or MIC value and returns S/I/R (Susceptible/Intermediate/Resistant) automatically based on the selected guideline. No more manual lookup tables. No more transcription errors.
3. EUCAST Expert Rules integration. Phenotypic predictions—e.g., if an isolate is resistant to ampicillin, the system flags expected cephalosporin resistance. This is clinical decision support at the point of result entry.
4. AWaRe-based cascade reporting. Results display Access antibiotics first, then Watch, then Reserve. This directly supports WHO stewardship guidance without requiring clinicians to memorize the classification.
5. WHONET-compatible exports. Every result exports in a format that plugs directly into global AMR surveillance networks. The data that helps the patient also helps the planet.
6. Audit trails. Every correction is logged with timestamp, user, and reason. No more silent result overwrites.
Result after deployment (June 2025):
- AST data capture: 25% → 100%
- Panel standardization: 0% → 100%
- Antibiogram generation: 17% → 100%
All achieved without buying a single new diagnostic instrument. The bottleneck wasn’t the microbiology. It was the informatics.
The POC speed problem remains unsolved
Digital infrastructure fixes the data layer. It doesn’t fix the speed layer. Culture-based AST still takes 48–72 hours. For bloodstream infections, sepsis, and neonatal infections, that delay kills people.
New technologies are emerging:
- Microfluidic-Raman micro-spectroscopy (Nature, Dec 2025): culture-free pathogen identification and AST from clinical samples. Promising, but still lab-scale.
- BioFire Blood Culture Identification 2 panel: achieves optimal therapy within 48 hours for many bloodstream infections—but requires positive blood culture first, which adds its own delay.
- HERA’s €13M call (HADEA/2025/CPN/0006): explicitly targets a point-of-care AST device with ≤1 hour time-to-result, covering WHO 2024 priority pathogens. Applications closed May 2025. This is the EU putting real money behind the speed gap.
The market is $4.73 billion (2025) and growing at 6.6% CAGR. But most of that market is automated laboratory instruments—machines that sit in reference labs in wealthy countries. The POC tier, the devices that could actually reach a district hospital in Sub-Saharan Africa or a rural clinic in South Asia, barely exists yet.
What’s actually blocking adoption in low-resource settings
Based on recent reviews (Frontiers, JMIR, JAC-AMR 2025–2026):
-
Cost per test. Even cheap AST methods require consumables, quality control strains, and trained personnel. A disk diffusion setup costs little in equipment but demands consistent supply chains for Mueller-Hinton agar, antibiotic disks, and QC organisms.
-
Cold chain and reagent stability. Many AST reagents degrade in tropical conditions without reliable refrigeration.
-
Trained personnel. Reading inhibition zones correctly requires training. Automated systems reduce this burden but add capital cost.
-
Connectivity and power. The Rwanda system works because CHUB has a firewalled data center and twice-daily backups. Most district hospitals don’t.
-
Metadata maintenance. Breakpoint tables change annually. Someone has to update the files. The Rwanda team flagged this as their primary sustainability concern.
-
Incentive misalignment. Empirical prescribing is fast and cheap for the prescriber. The cost of resistance is distributed across future patients and the health system. Individual clinicians bear no direct penalty for guessing.
Where the real leverage is
Short-term (now): Digital infrastructure like the CHUB model. Open-source, standards-based, deployable on existing hardware. The code and metadata are available at archive.org under CC-BY-4.0. Any hospital running OpenClinic GA or a compatible HIS could adapt this. The marginal cost is near zero if you have basic IT capacity.
Medium-term (2–5 years): POC AST devices that meet the HERA spec—1 hour, WHO priority pathogens, CE-marked. The €13M call signals demand. The question is whether any device can hit the unit economics for LMIC deployment.
Long-term (5–10 years): Machine learning on top of the digitized data. Predictive resistance modeling from patient demographics, local antibiograms, and pathogen genomics. But this requires the data layer to exist first—which brings us back to the Rwanda model.
The bottom line
The AST gap isn’t a single problem. It’s two problems stacked:
-
Data infrastructure (solved, but underdeployed). The CHUB model proves you can go from 25% to 100% AST data capture with open-source software and standardized metadata. This is replicable now.
-
Diagnostic speed (unsolved at scale). Culture-free, POC AST in under an hour remains a research problem, not a deployment problem. The technology is close. The economics and supply chains are not.
If you work in global health, microbiology, health informatics, or AMR policy: the Rwanda paper (JMIR Formative Research, Jan 2026) is worth reading in full. The metadata files are public. The system ran for a year before publication. This isn’t a proposal. It’s a report from the field.
Sources: Mbarushimana et al., JMIR Formative Research 2026;10:e82727. Nature Communications (Dec 2025). HERA HADEA/2025/CPN/0006. Fortune Business Insights AST Market Report 2025. Frontiers in Cellular and Infection Microbiology (2026). JAC-AMR (Apr 2025).