The ANA-negative patient. A patient he treated 15–20 years ago phoned from Bihar to say her ANA was now negative, so she believed she no longer had the disease.
The azathioprine patient. He stopped azathioprine in a patient whose disease was well controlled. Within a week, the patient came back with paraparesis (weakness in the legs), something they had never had before.

His question: were these decisions really "precise"? His centre has been working on precision medicine in lupus for the past 4–5 years to answer exactly this.

What precision medicine should deliver, as he defines it:

The right drug from the first visit. He wants a drug choice with about 95% efficacy for that specific patient, instead of today's success rate of roughly 60–65%.
Organ risk forecasting. Knowing at the first visit which organs are likely to be affected later.
Flare prediction. Knowing whether a patient will flare when medicines are tapered.
Smarter follow-up. A patient with mild, stable disease could be seen every six months instead of every two weeks or three months.
Damage prediction and prevention.
Cost-rational care. Choosing a drug that is effective and affordable, rather than trying drugs one after another.

2. Lupus (SLE) is not one disease

SLE varies between patients in its autoantibodies, clinical features, and immune profile.
Different patients' disease is driven by different processes: interferon, B cells, T cells, IL-17 or IL-1, NETosis (a type of white-cell death that releases inflammatory material) in some patients but not others, and metabolic abnormalities in others.
So precision care starts with working out which group the patient in front of you belongs to, and what is driving their disease.

What this mixing-together has cost the field:

Failed trials. Most lupus drug trials have missed their targets, unlike rheumatoid arthritis, where most trialled drugs succeed. Even rituximab, which doctors use every day, did not prove effective in its trials. He blames trials that lump different kinds of patients together and don't account for the background treatment patients are already on.
Problems in everyday care: steroid damage, kidney involvement (nephritis) picked up late, trial-and-error treatment order (hydroxychloroquine, then mycophenolate, then rituximab if those fail), too little evidence, and follow-up that can't be predicted.

3. The "endotype" idea

An endotype is a patient's underlying disease pattern. He believes most lupus patients keep the same one throughout their illness:
- A patient with kidney involvement usually keeps having kidney involvement.
- A patient with skin and mucosal disease usually stays that way.
- Some patients do switch, so the endotype can change, but most stay stable.
If the visible (clinical) pattern is stable, the underlying immune pattern is probably stable too.
A well-defined endotype should predict:
- which drug the patient will respond to, and
- which organs are likely to be affected over time.

4. The main obstacle: nine different ways of looking at lupus

Different specialists describe lupus in their own terms: immunologists (autoantibodies, immune cells), geneticists, transcriptomics researchers (gene activity), single-cell biologists, proteomics researchers, metabolism researchers, microbiome researchers, clinicians, and data analysts. The challenge is combining all of these views into one picture.

a) Grouping patients by autoantibodies

His team pooled data from 8–10 published studies worldwide and found six identifiable groups based on autoantibodies: anti-RNP, anti-Smith, anti-Ro (52/60) and La, antiphospholipid, and anti-dsDNA. This will be presented as a poster at APLAR.
Each group tends to affect different organs, so autoantibodies can partly predict organ involvement.
The groups overlap, though, so a second grouping by organ or clinical features is also needed.

b) Genetics and gene activity (transcriptomics)

Certain genes, such as those linked to interferon, can suggest risk of developing disease and which part of the immune system is altered.
Gene activity tests (for example "interferon-high" status or blood-based gene activity profiles) can help sort patients.
Caveat: these results aren't simply yes or no. A patient who tests "interferon-high" may also have other active pathways. This is one reason interferon-high patients don't reliably respond to anti-interferon drugs.

c) The integrated tool his team is building

They are building a tool that combines all available data, with modules for:

Cytokine biomarkers (inflammatory signalling molecules in the blood).
Complete blood count (CBC) and its ratios, used as a cheap stand-in for cytokine tests:
- Ratios such as the neutrophil-to-lymphocyte ratio (NLR) and the systemic immune-inflammation index (SII) track fairly well with ferritin and some cytokines.
- Where CBC ratios match the cytokine pattern, expensive full cytokine panels can be skipped for many patients.
Single-cell and spatial data, including:
- Urine used as a "liquid biopsy".
- Kidney biopsy results showing which cells are present.
- B-cell and T-cell analysis.
The tool uses whatever data is available for a patient and suggests which disease pathways could be targeted.

d) The biomarker problem

Hundreds of lupus biomarkers have been described, but most never went beyond two or three papers and were never tested in actual clinical use.
Testing hundreds of biomarkers per patient is too expensive, and he doubts anyone would fund it.
His shortcut: combining results across published studies, root-cause analysis, and "relational analysis." He acknowledges many statisticians disagree with relational analysis, but sees it as a way to avoid spending billions validating biomarkers that matter in only a few patients.
In practice, he focuses on simple, widely available tests: urine protein, urine tubular proteins, IL-6, IL-4, IL-8, and kidney biopsy findings. These are combined to predict which immune pathways are active.

e) Metabolism and the microbiome

Metabolic changes (such as mTOR pathway activation) vary from patient to patient. Existing drugs like metformin may help counter these, guided by simple blood and metabolic tests.
The gut microbiome also affects immune responses. That research is still developing, but he expects it to become part of prediction models eventually.

5. Choosing among B-cell treatments: an example of pathway-based choice

Current B-cell options:
- CD19 CAR-T cell therapy (deep B-cell removal), which has led to long-term remission in a small group of patients.
- Standard B-cell-depleting drugs targeting CD19 or CD20.
- Bispecific CD3×CD19 antibodies.
- Two-step approaches: removing B cells first, then using a BAFF inhibitor to stop them coming back.
Each option has its own cost, side effects, and limits. The question is which one suits which patient.
His key example, BAFF blockers:
- BAFF blockers worked in a good share of lupus patients, but failed in rheumatoid arthritis.
- The reason: lupus depends on a continuous supply of new self-attacking B cells, which BAFF blocking cuts off. In RA, memory B cells keep the disease going and cause flares, and BAFF blocking doesn't touch them.
- So knowing whether a patient's disease is driven by new B-cell recruitment, T-cell recruitment, or memory cells changes the treatment strategy completely.
This is why insights from all nine fields need to be built into one decision algorithm.

6. Matching drugs to pathways

The goal is pathway-targeted therapy. For each patient's endotype, find the dominant pathway. Possibilities include:
- type I interferon,
- dual-cytokine targeting,
- calcineurin inhibitors alone,
- intracellular signalling,
- complement-directed treatment,
- or several abnormal pathways at once.
Drug mapping (work in progress). For each drug, the team records the main pathway it blocks, plus the second- and third-level pathways affected. Two examples:
- Hydroxychloroquine acts mainly by blocking TLR7/9 and partly reduces interferon.
- Anifrolumab blocks the interferon pathway.
Why the knock-on effects matter. In some patients, blocking interferon alone may push the disease through other routes. This explains why some patients do very well on anti-interferon or anti-B-cell drugs while others get worse.
How the tool works:
- It first identifies which pathways are disrupted in the patient.
- It compares these with what each established drug blocks.
- It outputs a ranked list of the drugs best suited to that patient.

7. Treat-to-target, working step by step

The target is clear: DORIS remission where possible, or LLDAS (lupus low disease activity state) where remission isn't achievable. The open question is which pathway to target to get there.
A stepwise workup:
1. Clinical features, plus simple tests: autoantibody profile, complement levels, CBC ratios.
2. For harder cases: basic cytokine panels and cell proportions.
3. Then gene-activity markers to see which pathways are switched on.
4. Genetic data, where available.
The tool should predict which pathways are abnormal now and which may become abnormal once others are blocked, helping plan treatment. Its predictions will be updated as the science moves forward.

8. Digital tools and making them usable

Digital methods used: machine learning, digital reading of biopsies, registry data, and federated learning (training models across institutions without pooling patient data).
Usability is essential. Busy clinicians won't use a tool that takes more than about 10 minutes.
- Even the DAS28 score, which requires checking 28 joints, often gets skipped.
- A tool needing 100 inputs won't be used.
- The target is an answer within 10–15 minutes.
Better trial designs are needed to improve the tool:
- Biomarker-enriched trials (recruiting patients by biomarker).
- Basket/platform trials (testing one drug across related diseases, since drugs and diseases overlap).
- Adaptive and response-adaptive designs.
- Withdrawal, escalation, and de-escalation studies.

9. Gaps that still need closing

Under-represented populations. There is little genetic, cytokine, or interferon data on Indian / other Asian populations. Most data comes from white (Caucasian) patients.
Lab tests aren't standardised. Autoantibody results, especially blot-based tests, differ between labs, and reporting isn't standardised worldwide.
Cost and access limit testing.
Validation is still needed.
Long-term patient data is scarce.

10. Roadmap

His team's plan has four stages:

Characterise the disease.
Classify it.
Predict its course.
Plan how and when to intervene.

They are currently at stage two: grouping patients into clusters. Early prediction work uses existing data, and drug-intervention data comes from a database mapping the pathways each drug blocks. He notes the design draws on key landmark trials and studies.

He closed by inviting the audience to PRISM, a conference in Bangalore on September 26–27. It focused on precision medicine and personalised care in autoimmune diseases, especially rheumatoid arthritis, lupus, scleroderma, and spondyloarthropathy.

Key takeaways

Lupus is many diseases under one name. Precision care begins by working out each patient's endotype, which usually stays stable over time.
A single algorithm should combine autoantibody clusters, clinical patterns, cheap CBC ratios, selected biomarkers, gene activity, genetics, metabolism, and eventually the microbiome.
Matching a patient's disrupted pathways to what each drug blocks (including its knock-on effects) can produce a ranked drug list. This replaces trial-and-error treatment.
The BAFF example shows why the mechanism matters: the same drug class can work in lupus and fail in RA.
To be used in practice, the tool must be fast (10–15 minutes), affordable, and validated in Indian and other under-represented populations.