Nephrolytics
Reimagining kidney care intelligence. From five disconnected portals to one unified, predictive, human-readable platform.
Chronic kidney disease affects over 850 million people worldwide. I led research and design across a 12-week pilot spanning three dialysis centers — turning 34 stakeholder interviews into a platform nephrologists, coordinators, and patients each trust.
Built by the person who lived the problem
Nephrolytics began with twenty years of clinical frustration. Dr. Naeem Raheem spent his career watching critical patient information scatter across portals that were never built for the way nephrology actually works. Alongside his brother, he set out to build the opposite: the first AI-powered assistant built for nephrology, by a nephrologist.
“Having a product created by a nephrologist does make a difference.”
My first work on this project was sitting with his product demo — a real patient case, not a script. That walkthrough became the seed of everything that followed: before a single Stage 3 CKD patient could sit down with his nephrologist, five separate documents had to be manually reconciled. No synthesis. No trend context. This wasn’t a technology gap — it was a patient-safety gap.
- A multi-page Epic discharge summary
- A multi-page infectious disease progress note
- A handwritten clinical note
- PDF lab results, unparseable by the EHR
- An office note from the last clinic visit
“Clinicians have no unified, real-time view of a CKD patient’s progression — forced to piece it together across disconnected systems before the conversation even starts.”
Operational inefficiencies compounding the problem
Beyond the five-document problem, day-to-day workflow research surfaced a set of recurring operational inefficiencies. These include:
- Difficulty finding and accessing relevant patient data
- Managing multiple passwords for different EMRs
- Frequent duplication of similar information
- A lack of seamless data transfer between systems, especially for dialysis patients receiving treatment from multiple hospitals
- Inefficiencies in data entry and transfer leading to the recreation of information (mainly SOAP notes) from scratch
- Lab results that are usually scanned documents and are non-editable
- Frustrations with EMRs not communicating seamlessly with pharmacies, labs, and other specialists
Challenges related to data gathering from nephrologists and providers, as well as the cumbersome nature of reviewing data and making projections manually, further add to the complexity of managing resources.
| User Group | Core Problem | Frequency | Clinical Impact |
|---|---|---|---|
| Nephrologists |
| Every visit | 18+ min reconciliation/visit; 41% unplanned dialysis initiations |
| Care Coordinators |
| Daily | 8 hrs/week manual reporting; 5.2-day alert latency |
| Patients |
| Every lab cycle | 67% can’t interpret their eGFR; 2.3× higher hospitalization (non-adherent) |
| IT / Informatics |
| Every new integration | 6.2-week avg. new vendor integration |
Research and Key Insight
I designed a mixed-method research plan spanning nephrologists, care coordinators, patients, and IT leads across India, the US, and the UK — pairing what people said with what I watched them actually do.
A day in the nephrologist’s chair
From the 7-day diary study, we mapped the emotional experience of each user type across a typical week. The low points below run stepwise, in the order they occur.
What 34 conversations actually revealed
Themes synthesized from the interviews, enriched with direct observations from Dr. Naeem Raheem’s own product demo.
The Five-System Problem
Every clinician accessed 3+ separate systems per patient review. In the demo, a single patient required an Epic discharge summary, an ID progress note, a handwritten note, PDF labs from LabCorp, and a prior office note — all before the patient sat down.
“These are the labs that came from the hospital in a format which is PDF and not direct connection.”
The Virtual Fellow Gap
Clinicians wanted something that functions like a knowledgeable junior colleague who has already reviewed the notes before rounds — an overview, not another portal to search.
“We wanted to create an application that works like a virtual fellow, giving you overview, updates, last visit summaries, lab trends.”
Documentation Kills Clinical Time
Documentation burden was the single most-cited threat to clinical quality — time that should belong to the patient instead goes into typing.
“We’ll start talking to the patient in a natural manner and continue to hit all the things we want to deal with, with a normal conversation.”
Gap-in-Care Goes Undetected
Clinicians regularly missed guideline-recommended actions — not from negligence, but from information overload with no system flagging the gap.
“SMART Recommendation mentions no recent iron studies were found — and shows what source was used to give that recommendation.”
Alert Fatigue vs. Alert Absence
High-volume coordinators had disabled every automated alert due to noise; low-volume sites had none at all. Both were flying blind.
“If a patient’s creatinine usually is 3 and now it’s 4.4, it’ll pick it up. If it’s just 2, that would not be a smart alert.”
The Patient Comprehension Gap
10 of 12 patients could not correctly explain what eGFR measures without assistance, and 8 of 12 felt anxious receiving lab results but didn’t know what action to take. Patients with a family member in medical training reported dramatically better self-management confidence — a clear information-design opportunity.
“You also have the ability to generate a patient visit summary.”
The Feedback Loop Patients Want
Asked what would most help them manage their condition, the top answer was wanting to see a direct link between daily choices and lab outcomes — not generic advice, but personalized feedback tied to their own numbers.
“I want to know if what I’m eating is helping or hurting.”
Glimpses from user interviews
“Nephrolytics isn’t just an AI scribe. It is an intelligent copilot built to support the way nephrologists practice every day.”
“Having a product created by a nephrologist does make a difference.”
“We have the data. It just lives in five places that don’t talk to each other.”
“If I had one screen showing eGFR trend, medication history, and last dialysis adequacy — my consult time would be cut in half.”
“These are the labs from the hospital in PDF — not a direct connection. We need to parse, digitize, and quantify them.”
“I don’t need 50 alerts. I need three: who missed a session today, who is fluid-overloaded, who hasn’t had labs in 60 days.”
“Every time I get a report it says ‘Creatinine: 4.2’. Is that good? Is that bad? Nobody tells me.”
“I track my fluid intake in a notebook. My doctor has never seen that notebook.”
Three people, three very different definitions of "it’s working"
Curated from 34 interviews down to the three roles whose success criteria pulled hardest against each other.
I spend 40% of my clinical time chasing lab results across five different portals.
- One trended, annotated view per patient at point of care
- Flag rapid CKD progression before dialysis eligibility
No predictive risk scoring, and dialysis logs still get re-typed by hand from separate vendor portals.
My patients miss sessions and I find out days later — I need real-time data, not a weekly CSV dump.
- Same-day visibility into attendance and fluid compliance
- Compliance reports in minutes, not hours
Six to ten hours a week lost to manual CSV exports before a single patient gets called.
My doctor talks about creatinine and I nod — I just want to know: am I getting better or worse?
- Lab results in plain language, with a clear trend
- See how her diet actually connects to her numbers
Anxiety spikes between appointments with no data to anchor how she’s actually doing.
What they’re trying to do
Three roles, three different jobs to be done — but each success metric had to be provable, not aspirational.
See the full kidney health story of a patient in under 60 seconds at point of care.
Identify high-risk patients before they cross the dialysis threshold.
Reconciliation < 5 min; 0 missed stage transitions
Know, in real time, which patients need outreach today — without manual checking.
Produce compliance reports in minutes, not hours.
Alert latency < 2 hrs; reports < 15 min
Understand what their labs mean and whether they are improving.
See the direct impact of diet and lifestyle choices on their kidneys.
Comprehension score > 7/10
Four patterns that showed up in every interview
Every card below appeared, unprompted, across more than half of that role’s interviews.
The five-portal problem
Every clinician interviewed accessed 3+ separate systems to complete one patient review. Context-switching was the single biggest workflow burden they named.
“We have the data. It just lives in five places that don’t talk to each other.”
Reactive, not predictive
Clinicians had no systematic way to flag high-risk patients — risk surfaced through intuition, manual chart review, or an ER visit.
“By the time I know a patient is in trouble, the window for intervention is often gone.”
The CSV graveyard
Every coordinator described the same Monday ritual: downloading attendance CSVs, cross-referencing schedules, hand-building compliance reports in Excel.
“I spend Monday morning building last week’s report instead of calling the patients who missed sessions.”
The comprehension gap
Patients could not explain what their own eGFR measured without help. Nearly all felt anxious about results with no idea what action to take.
“Every report says “Creatinine: 4.2 mg/dL.” Is that good? Is that bad? Nobody tells me.”
What the platform had to do
Every need statement below is traceable back to specific interviews — priority wasn’t a guess.
A single, time-ordered patient record spanning labs, imaging, dialysis logs, and medications — queryable by date, biomarker, or event.
ML-powered risk scores updated on every lab refresh, surfacing patients near Stage 5 or dialysis eligibility with explainable factors.
Push notifications for missed sessions, fluid overload, and medication non-adherence within 2 hours of the event.
Clinical lab values translated into plain-language summaries with trend indicators and dietary context.
HL7 FHIR R4-compliant ingestion across Epic, Cerner, Meditech, and major lab vendors, under 4-hour latency.
Auto-generated weekly and monthly compliance reports, exportable to PDF/Excel, eliminating manual reconciliation.
Five principles, held consistently across every feature decision
Consolidate before you complicate
Every feature starts from a data unification layer — no analytics on top of fragmented data.
Alert precision over volume
Fewer, higher-confidence signals. Coordinators set their own thresholds per patient.
Translate for the audience
Clinicians see biomarker charts. Patients see “your kidney health improved this month.” Same data.
FHIR or nothing
Every integration is standards-based. No bespoke ETL, no vendor lock-in on data exchange.
Explainable AI
Every risk score shows its contributing factors in plain language. Clinicians can override or annotate.
Key features of the product
Schedule Patient Page
- •List view: lets providers search for patients by name, ID, or date of birth as an alternative to the card view
- •Providers can view all scheduled patients for the day and visit their previous visit summary
Individual Patient Cards View
- •Card view (home screen): auto-populates patient cards by pulling data directly from connected EHRs, eliminating manual entry
Patient Intake form
- •Displays the patient's photo (if uploaded), name, DOB, age, patient ID, assigned primary clinician, and account status at the top of the page for instant identification
- •A chronological timeline of all past and upcoming appointments with date, clinician, appointment type, and outcome — each entry links to the corresponding consultation note
- •Clinicians attach SOAP notes, outcome measures, and progress reports directly to the patient record; notes are version-controlled and cannot be deleted, only amended with an audit trail
- •Stores referral letters, imaging reports, pathology results, and signed consent forms; files can be uploaded by staff or received via secure inbox integration
- •An active treatment plan with goals, planned sessions, and progress markers; clinicians update goal achievement status after each consultation
- •Shows the patient's account balance, invoice history, payment method on file, and Medicare claim status; staff can raise a new invoice or process a rebate from this panel
Patient Details Page
A single patient view that functions like a virtual fellow — surfacing everything a nephrologist needs before walking into the room.
- •Patient summary panel (left): overview, last visit summary, lab trends, and outside encounters (admissions, discharges, other care events)
- •Last visit summary: recaps what was addressed previously (e.g., stable renal function, antibiotic course, deferred procedures) for continuity before the visit
- •Linked source documents: aggregates original discharge summaries, ID notes, handwritten notes, and lab PDFs for reference
- •Document summarization: condenses linked documents, including PDF-only labs, into readable summaries
- •Lab digitization & table view: converts PDF labs into structured tables, flags abnormal values, compares against the last 1–2 results, and includes transplant-specific labs when relevant
- •Medications & documents sections kept accessible alongside the summary view
- •Vitals/labs trending (top right): trend vitals and individual labs by search
- •Smart Recommendations panel: flags care gaps (e.g., missing iron studies) with source guideline citations; providers accept, edit, or defer; customizable for RPM/CCM enrollment, research protocols, and CKCC/advanced kidney disease tracking
Interim Screen:
- •Generated after visit recording ends, reflecting the conversation captured during the visit
- •Color-coded recommendation tracking: addressed items turn green; patient refusals and deferred/flagged items are shown in distinct colors for quick scanning
- •Auto-captured coding: diagnoses populate with ICD-10 codes, status, and plan — all editable
- •CPT code generation: calculated from the conversation, factoring in MDM complexity
- •Labs, referrals, and imaging orders: anything mentioned in the visit appears here, with labs capable of auto-routing to the lab system via EHR integration
End Note
- •One-click generation: pressing "Accept" on the interim screen produces the final note
- •Bottom-line-up-front SOAP format: leads with the most actionable information rather than burying it
- •Action summary section: highlights medication changes, ordered labs, follow-up plan, and billing at the top for quick review
- •Assessment & plan: included for full documentation reference
- •Copy function: one button copies everything from the action items through the plan (and separately, reason-for-visit through physical exam) for EHRs without direct push integration
- •Sign-off: provider signs the note once reviewed
- •Patient visit summary: can be generated from the finalized action plan for patient-facing use
How far we have come
Each card tells the story: where things stood before Nephrolytics, and where they stand now.
| Metric | Before | After | Change |
|---|---|---|---|
| Data reconciliation time per visit | 18 min | 4 min | −78% |
| Unplanned dialysis initiations | 41% | 18% | −56% |
| Weekly coordinator reporting time | 8.0 hrs | 0.75 hrs | −87% |
| Lab-to-clinician-review latency | 5.2 days | 1.1 hrs | −99% |
| Patients unable to interpret eGFR | 67% | 31% | −54% |
| Patient comprehension score (0–10) | 3.2 | 7.6 | +4.4 pts |
| Coordinator alert fatigue (0–10) | 7.8 | 2.9 | −63% |
| Integration go-live, new vendor | 6.2 wks | 3.5 days | −92% |
