Introduction
This document accompanies the keynote Medicine × LLM × R and
shows, step by step, how selected medical tasks from the talk can be
carried out with large language models (LLMs) in R. The examples
progress from basic inference to retrieval-augmented generation (RAG)
and a simple application:
| 1 |
Patient communication, translation |
rollama |
Explain laboratory results in simple Malay |
| 2 |
Triage / text classification |
rollama |
Zero-shot classification of complaints |
| 3 |
Information extraction |
tidyllm |
Structure clinical notes into a data frame |
| 4 |
Extraction and evaluation |
ellmer |
Extract variables from notes, evaluate against a gold
standard |
| 5 |
Decision support (calculation) |
ellmer |
Tool calling: R computes eGFR, the LLM explains |
| 6 |
Patient education, grounded in a document |
ragnar |
RAG over a Malay Wikipedia page |
| 7 |
Simple application |
ellmer + shinychat +
ragnar |
Patient-education chatbot |
All clinical notes are synthetic. The code is for
demonstration only and is not intended for clinical use.
Requirements
Ollama, installed and
running, with the following models:
ollama pull medgemma1.5:4b # chat model, Sections 1-4
ollama pull qwen3.5:4b # chat model with tool calling, Section 5
ollama pull bge-m3 # multilingual embedding model, Sections 6-7
ollama pull hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0 # Malay-capable chat model, Sections 6-7
R packages:
install.packages(c("rollama", "tidyllm", "ellmer", "ragnar", "jsonlite",
"shiny", "shinychat", "bslib"))
MJMSTable is also used for the descriptive table in
Section 3.
From the talk’s repository: the synthetic notes in
data/clinical_notes.csv, and the helper functions
clean_thought() and extract_num() in
functions.R.
The LLM outputs below were generated when this document was compiled;
they will differ between models, model versions and runs.
model = "medgemma1.5:4b" # chat model used throughout
Sections 1–4 use MedGemma, an open-weight medical model (Sellergren et al. 2025; Google Research and Google DeepMind
2026).
1. rollama: basic local inference
rollama sends prompts to models running locally in
Ollama (Gruber and Weber
2025). Nothing leaves the computer, which matters for patient
data. A fixed seed and temperature = 0 make
the output more reproducible. clean_thought() (in
functions.R) removes the model’s reasoning text, keeping
only the answer.
library(rollama)
ping_ollama() # is Ollama running?
# pull_model(model) # download once
finding = "HbA1c 8.9%, fasting glucose 11.2 mmol/L, BP 150/95 mmHg."
query(
paste("Explain these results to a patient in simple Bahasa Melayu,",
"in 3 sentences. Do not give a diagnosis.", finding),
model = model,
output = "data.frame", screen = FALSE,
model_params = list(seed = 2026, temperature = 0)
) -> res
clean_thought(res$response) |> cat()
#> Keputusan ujian menunjukkan gula darah anda tinggi dan tekanan darah anda juga tinggi. Ini adalah perkara penting yang perlu kita kawal untuk kesihatan anda. Mari kita bincang langkah selanjutnya untuk menguruskan ini.
2. rollama: zero-shot triage classification
make_query() builds a structured prompt from the text,
the task and a system role. The result is an ordinary R vector, so the
usual tools for diagnostic accuracy and agreement apply.
complaints = c(
"Crushing chest pain radiating to the left arm, sweating, 30 minutes.",
"Mild sore throat for two days, no fever, eating well.",
"Sudden weakness of the right arm and slurred speech since this morning."
)
q = make_query(
text = complaints,
prompt = "Classify the triage urgency: 'emergency', 'urgent' or 'non-urgent'.",
system = "You are a triage assistant. Answer with one category only, in lowercase."
)
triage = query(q, model = model, output = "text", screen = FALSE,
model_params = list(seed = 2026, temperature = 0)) |> clean_thought()
triage_df = data.frame(complaint = complaints, triage = triage)
triage_df
Comparison with a reference classification (here, a hypothetical
triage nurse):
nurse = c("emergency", "non-urgent", "emergency")
table(LLM = triage_df$triage, Nurse = nurse)
#> Nurse
#> LLM emergency non-urgent
#> emergency 2 0
#> non-urgent 0 1
3. tidyllm: structuring clinical notes
tidyllm defines the information to extract as a schema;
the model is constrained to return JSON that follows it (Brüll 2026; Arifin 2026).
library(tidyllm)
library(jsonlite)
patient_schema = tidyllm_schema(
name = "PatientHistory",
patient_name = field_chr("Patient's name; empty if not stated"),
age = field_dbl("Age in years. If given in months, convert to a fraction of a year."),
gender = field_fct("Gender of the patient", .levels = c("male", "female")),
symptoms = field_chr("Current symptoms reported by the patient", .vector = TRUE)
)
One note:
raw_text = "Mrs. Amy, a 35-year-old lady came to the outpatient clinic with the
complaint of fever for three days and cough."
result = llm_message(paste("Extract the patient's note according to the schema.", raw_text)) |>
chat(.provider = ollama(), .model = model, .json_schema = patient_schema,
.temperature = 0, .seed = 2026)
# Cloud alternative: chat(.provider = gemini(), .model = "gemini-3.1-flash-lite", ...)
get_reply(result) |> minify()
#> {"patient_name":"Mrs. Amy","age":35,"gender":"female","symptoms":["fever","cough"]}
Several notes, saved as JSON Lines and then read into a data
frame:
notes = read.csv("data/clinical_notes.csv")
output_file = "pt_data.jsonl"
if (file.exists(output_file)) file.remove(output_file) # start a new file on each run
#> [1] TRUE
for (note in notes$note) {
tryCatch({
llm_message(paste("Extract the patient's note according to the schema.", note)) |>
chat(.provider = ollama(), .model = model, .json_schema = patient_schema,
.temperature = 0, .seed = 2026) |>
get_reply() |> minify() |>
cat("\n", file = output_file, append = TRUE, sep = "")
}, error = function(e) cat("Error:", conditionMessage(e), "\n"))
}
patient_data = stream_in(file(output_file), verbose = FALSE) # read the JSON Lines file
patient_data
From unstructured text to the descriptive table we already know.
MJMSTable::descriptive_tbl(patient_data, included_var = c("age", "gender"))
Table : Patient Demographics (n = 8).
| Variables |
Total
n (%) |
| age |
52.4 (15.6) |
| gender |
|
| female |
4 (50.0) |
| male |
4 (50.0) |
4. ellmer: extraction and evaluation
ellmer provides one interface to many providers, cloud
and local (Wickham et al.
2026). The schema is written with R types, and
parallel_chat_structured() returns one row per note. For
the numeric variables, the model copies the relevant text from the note
(e.g. '132 umol/L'), or an empty string if the value is not
reported.
The synthetic notes contain deliberate traps: an ex-smoker, a note
without smoking status, missing laboratory results and “percent” written
in words.
notes = read.csv("data/clinical_notes.csv")
notes[, c("id", "note")]
library(ellmer)
type_patient = type_object(
age = type_number("Age in years, as written in the note"),
sex = type_enum(values = c("male", "female"),
description = "Sex of the patient, from words such as man, lady, Encik, Puan"),
bp_text = type_string("Blood pressure value and unit only, copied exactly from the note, e.g. '150/95 mmHg'. Empty string if not reported."),
hba1c_text = type_string("HbA1c value and unit only, copied exactly from the note, e.g. '8.9%'. Empty string if not reported."),
creatinine_text = type_string("Serum creatinine value and unit only, copied exactly from the note, e.g. '132 umol/L'. Empty string if not reported."),
smoker = type_enum(values = c("yes", "no", "unknown"),
description = "Is the patient a CURRENT smoker? Ex-smokers and never-smokers are 'no'. 'unknown' if smoking is not mentioned.")
)
chat = chat_ollama(
model = model,
system_prompt = "You extract data from clinical notes. Do not guess values that are not stated.",
params = params(temperature = 0, seed = 2026)
)
extracted = parallel_chat_structured(chat, as.list(notes$note), type = type_patient)
extracted
A small model often returns wrong or missing numbers when asked for
them directly, and may fill in typical values that are not in the note.
The LLM is therefore used only to locate the text, and R extracts the
numbers from it with extract_num() (in
functions.R). Text without a value, e.g. “No HbA1c done
this visit”, gives NA.
extracted$sbp = extract_num(extracted$bp_text, 1)
extracted$dbp = extract_num(extracted$bp_text, 2)
extracted$hba1c = extract_num(extracted$hba1c_text)
extracted$creatinine = extract_num(extracted$creatinine_text)
extracted
Evaluation against the gold standard
The gold standard is stored alongside the notes. Agreement is defined
as equal values, or missing in both.
gold = notes[, c("age", "sex", "sbp", "dbp", "hba1c", "creatinine", "smoker")]
agree = function(x, y) (is.na(x) & is.na(y)) | (!is.na(x) & !is.na(y) & x == y)
Field-level accuracy (%):
acc = sapply(names(gold), function(x) mean(agree(extracted[[x]], gold[[x]])))
round(acc * 100, 2)
#> age sex sbp dbp hba1c creatinine smoker
#> 100 100 100 100 100 100 100
For systematic evaluation across tasks, models and repeated runs, see
vitals (Couch
2025).
5. ellmer: tool calling, let R do the maths
LLMs are unreliable calculators. An R function is exact and can be
validated; the LLM decides when to call it and explains
the result. The function below implements the race-free CKD-EPI 2021
creatinine equation (Inker et al.
2021).
egfr_ckdepi2021 = function(creatinine_umol, age, sex) {
scr = creatinine_umol / 88.4 # umol/L -> mg/dL
female = sex == "female"
k = ifelse(female, 0.7, 0.9)
a = ifelse(female, -0.241, -0.302)
142 * pmin(scr / k, 1)^a * pmax(scr / k, 1)^-1.200 *
0.9938^age * ifelse(female, 1.012, 1)
}
egfr_ckdepi2021(132, age = 64, sex = "female") |> round(1)
#> [1] 38.9
tool_egfr = tool(
egfr_ckdepi2021,
name = "egfr_ckdepi2021",
description = "Estimate GFR (mL/min/1.73 m2) using the race-free CKD-EPI 2021 creatinine equation.",
arguments = list(
creatinine_umol = type_number("Serum creatinine in umol/L"),
age = type_number("Age in years"),
sex = type_enum(values = c("male", "female"), description = "Sex of the patient")
)
)
Here we use qwen3.5:4b, because medgemma1.5 does not support tool
calling. The answer should give the KDIGO CKD GFR category (Kidney Disease: Improving
Global Outcomes (KDIGO) CKD Work Group 2013): an eGFR of 38.9
mL/min/1.73 m² is category G3b, moderately to severely decreased.
model_tool = "qwen3.5:4b"
# temperature > 0 for a thinking model: at temperature = 0 it can get stuck in a
# thinking loop; the fixed seed keeps the run reproducible
chat = chat_ollama(model = model_tool, params = params(temperature = 0.5, seed = 2026))
chat$register_tool(tool_egfr)
t0 = Sys.time()
res_tool = chat$chat("Puan Aminah is a 64-year-old woman with serum creatinine 132 umol/L.
What is her eGFR, which KDIGO CKD GFR category and
which clinical term is it?", echo = "none")
time_tool = as.numeric(difftime(Sys.time(), t0, units = "secs"))
res_tool
#> Based on the calculations using the race-free CKD-EPI 2021 creatinine equation:
#>
#> **eGFR:** 38.9 mL/min/1.73 m² (rounded from 38.8845)
#>
#> **KDIGO CKD GFR Category:** Stage 3b
#> - Range: 30–44 mL/min/1.73 m²
#>
#> **Clinical Term:** Moderate to severe reduction in kidney function (CKD Stage
#> 3b)
#>
#> This indicates that Puan Aminah has moderate-to-severe chronic kidney disease,
#> which requires appropriate clinical management and monitoring according to
#> KDIGO guidelines for CKD.
cat("Time taken (with tool):", round(time_tool, 1), "seconds\n")
#> Time taken (with tool): 7.2 seconds
For comparison, the same question without the tool. If you try this
out, you will notice that qwen3.5:4b takes considerably longer, as it
reasons through the calculation itself.
chat_no_tool = chat_ollama(model = model_tool, params = params(temperature = 0.5, seed = 2026))
t0 = Sys.time()
res_no_tool = chat_no_tool$chat("Puan Aminah is a 64-year-old woman with serum creatinine 132 umol/L.
Calculate her eGFR with CKD-EPI 2021 creatinine equation.
Which KDIGO CKD GFR category and which clinical term is it?",
echo = "none")
time_no_tool = as.numeric(difftime(Sys.time(), t0, units = "secs"))
res_no_tool
#> Here is the step-by-step calculation and determination of Puan Aminah's kidney
#> function status based on the provided data.
#>
#> ### 1. Calculation of eGFR (CKD-EPI 2021 Equation)
#>
#> **Patient Data:**
#> * **Gender:** Female
#> * **Age:** 64 years
#> * **Serum Creatinine ($SCr$):** $132 \, \mu\text{mol/L}$
#>
#> **Step A: Convert Serum Creatinine to mg/dL (if required by formula)**
#> The CKD-EPI equation typically uses mg/dL for the ratio calculation.
#> $$ \text{Conversion Factor} = 0.0113 $$
#> $$ SCr (\text{mg/dL}) = 132 \times 0.0113 \approx 1.49 \, \text{mg/dL} $$
#>
#> **Step B: Determine the Threshold ($\kappa$)**
#> For females in the CKD-EPI 2021 equation, the threshold $\kappa$ is **$0.71 \,
#> \text{mg/dL}$** (or $63 \, \mu\text{mol/L}$).
#>
#> **Step C: Apply the Equation**
#> Since $1.49 \, \text{mg/dL} > 0.71 \, \text{mg/dL}$, we use the equation for
#> creatinine levels above the threshold:
#> $$ eGFR = 142 \times \left( \frac{SCr}{\kappa} \right)^{-0.203} \times
#> (0.9938)^{\text{Age}} $$
#>
#> * **Ratio ($SCr / \kappa$):** $1.49 / 0.71 \approx 2.10$
#> * **Creatinine Term:** $(2.10)^{-0.203} \approx 0.86$
#> * **Age Term:** $(0.9938)^{64} \approx 0.67$
#>
#> $$ eGFR = 142 \times 0.86 \times 0.67 $$
#> $$ eGFR \approx 81 \, \text{mL/min/1.73m}^2 $$
#>
#> *Note: Depending on the specific calculator implementation (e.g., MDCalc vs.
#> manual formula), values may slightly vary between **70–80 mL/min/1.73m²**.
#> Standard clinical calculators often yield approximately **70 mL/min/1.73m²**
#> for this input profile.*
#>
#> We will use the standard calculator consensus value of **70 mL/min/1.73m²** for
#> classification purposes, as it is more commonly cited in clinical guidelines
#> for this creatinine level.
#>
#> ### 2. KDIGO CKD GFR Category
#>
#> According to the KDIGO (Kidney Disease: Improving Global Outcomes)
#> classification system based on eGFR:
#>
#> * **G1:** $\geq 90$ mL/min/1.73m²
#> * **G2:** $60 - 89$ mL/min/1.73m²
#> * **G3a:** $45 - 59$ mL/min/1.73m²
#> * **G3b:** $30 - 44$ mL/min/1.73m²
#> * **G4:** $15 - 29$ mL/min/1.73m²
#> * **G5:** $< 15$ mL/min/1.73m²
#>
#> **Puan Aminah's Category:**
#> With an eGFR of approximately **70 mL/min/1.73m²**, she falls into the **G2**
#> category (60–89 mL/min/1.73m²).
#>
#> ### 3. Clinical Term
#>
#> The clinical term corresponding to CKD GFR Category **G2** is:
#> **CKD Stage 2** (or "Mildly decreased GFR").
#>
#> ***
#>
#> ### Summary Answer
#>
#> * **eGFR:** Approximately **70 mL/min/1.73m²** (Range 60–89).
#> * **KDIGO GFR Category:** **G2**
#> * **Clinical Term:** **CKD Stage 2**
cat("Time taken (without tool):", round(time_no_tool, 1), "seconds\n")
#> Time taken (without tool): 93.3 seconds
Time and tokens used, with and without the tool:
tok_tool = chat$get_tokens()
tok_no_tool = chat_no_tool$get_tokens()
data.frame(
run = c("With tool", "Without tool"),
seconds = round(c(time_tool, time_no_tool), 1),
assistant_turns = c(nrow(tok_tool), nrow(tok_no_tool)),
input_tokens = c(sum(tok_tool$input, na.rm = TRUE), sum(tok_no_tool$input, na.rm = TRUE)),
output_tokens = c(sum(tok_tool$output, na.rm = TRUE), sum(tok_no_tool$output, na.rm = TRUE))
)
6. ragnar: grounding answers in a document (RAG)
Retrieval-augmented generation (RAG) first retrieves relevant
passages from a document store, then lets the LLM answer from them. This
reduces hallucination and allows the answer to cite its source (Kalinowski and Falbel
2026). Here the store is built from the Malay Wikipedia page
Kencing manis.
read_as_markdown() cannot read the Wikipedia URL
directly, because Wikipedia rejects the default User-Agent of the Python
library it uses (HTTP 403). Instead, the article is downloaded as plain
text through the Wikipedia API with download.file(), which
also leaves out the page menus and sidebars, and saved as a markdown
file for read_as_markdown().
library(ragnar)
# Plain-text article from the Wikipedia API (no menus or sidebars)
url = "https://ms.wikipedia.org/w/api.php?action=query&prop=extracts&explaintext=1&redirects=1&format=json&titles=Kencing_manis"
json_file = "kencing_manis.json"
md_file = "kencing_manis.md"
download.file(url, destfile = json_file, mode = "wb")
writeLines(jsonlite::fromJSON(json_file)$query$pages[[1]]$extract,
md_file, useBytes = TRUE)
The store is a DuckDB file. It should be on a local disk:
cloud-synced folders (e.g. pCloud Drive) can break DuckDB’s temporary
write-ahead (.wal) file.
store_location = path.expand("~/ragnar_store/kencing_manis.duckdb") # local disk
dir.create(dirname(store_location), showWarnings = FALSE)
store = ragnar_store_create(
store_location,
embed = \(x) embed_ollama(x, model = "bge-m3"), # multilingual embeddings
overwrite = TRUE # rebuild from scratch
)
chunks = read_as_markdown(md_file) |> markdown_chunk()
chunks
ragnar_store_insert(store, chunks)
ragnar_store_build_index(store)
Retrieval of the most relevant chunks for a question:
ragnar_retrieve(store, "Apakah tanda-tanda kencing manis?", top_k = 3)
The store is registered as a tool, so the LLM searches it before
answering. Here we use Apertus-SEA-LION-v4-8B
from AI Singapore, a model that supports Bahasa Melayu:
model_my = "hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0"
chat = chat_ollama(
model = model_my,
system_prompt = paste(
"You explain diabetes to patients in simple Malay or English.",
"Search the knowledge store before answering and cite it.",
"You do not diagnose. Always advise seeing a doctor."
),
params = params(temperature = 0, seed = 2026)
)
ragnar_register_tool_retrieve(chat, store, top_k = 5)
chat$chat("Apakah tanda-tanda awal kencing manis?")
#> Tanda-tanda awal kencing manis (diabetes) termasuk:
#>
#> 1. **Kencing kerap (polyuria)** - Anda mungkin perlu ke tandas lebih kerap,
#> terutamanya pada waktu malam.
#> 2. **Dahaga yang berlebihan (polydipsia)** - Anda mungkin merasa sangat dahaga
#> dan minum lebih banyak air daripada biasa.
#> 3. **Kelaparan yang berlebihan (polyphagia)** - Anda mungkin merasa lapar lebih
#> kerap dan makan lebih banyak daripada biasa.
#> 4. **Penurunan berat badan yang tidak disengajakan** - Anda mungkin kehilangan
#> berat badan walaupun anda makan seperti biasa atau lebih banyak.
#> 5. **Kelelahan atau keletihan yang berlebihan** - Anda mungkin merasa sangat
#> letih atau lemah.
#> 6. **Penglihatan kabur** - Anda mungkin mengalami penglihatan yang kabur atau
#> tidak jelas.
#> 7. **Luka yang lambat sembuh** - Luka atau calar pada kulit anda mungkin
#> mengambil masa yang lebih lama untuk sembuh.
#> 8. **Kulit yang gatal atau kering** - Anda mungkin mengalami kulit yang gatal
#> atau kering, terutamanya di kawasan seperti kaki atau tangan.
#>
#> Jika anda mengalami tanda-tanda ini, adalah penting untuk berjumpa dengan
#> doktor untuk pemeriksaan dan diagnosis yang tepat.
The same steps apply to other documents, for example a local clinical
practice guideline in PDF:
read_as_markdown("path/to/cpg.pdf").
7. ellmer + shinychat +
ragnar: a patient-education app
A Shiny app cannot run inside a static HTML document. Run the code
below in R (or code/06_ellmer_shinychat_ragnar_app.R) after
building the store in Section 6 (Cheng et al. 2026).
library(shiny)
library(shinychat)
library(ellmer)
library(ragnar)
store_location = path.expand("~/ragnar_store/kencing_manis.duckdb") # as in Section 6
store = ragnar_store_connect(store_location, read_only = TRUE)
ui = bslib::page_fillable(
bslib::card(
bslib::card_header("Pembantu pendidikan pesakit: kencing manis (demo)"),
chat_ui("chat")
)
)
model_my = "hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0"
server = function(input, output, session) {
chat = chat_ollama(
model = model_my,
system_prompt = "You explain diabetes to patients in simple Malay or English.
Search the knowledge store before answering and cite it.
You do not diagnose. Always advise seeing a doctor."
)
ragnar_register_tool_retrieve(chat, store, top_k = 5)
chat_server("chat", chat, history = FALSE)
}
shinyApp(ui, server)
For a quick personal chat in the browser, without writing an app:
chat = chat_ollama(model = model_my)
ragnar_register_tool_retrieve(chat, store)
live_browser(chat)
Not a virtual doctor. Keep a human in the loop and
validate any tool before clinical use. Wikipedia is used here only as an
openly licensed example; for practice, use authoritative sources such as
clinical practice guidelines.
References
Arifin, Wan Nor. 2026.
Structuring Clinical Notes into Analyzable
Data with R and LLMs.
https://wnarifin.github.io/workshop.html.
Brüll, Eduard. 2026.
Tidyllm: Tidy Integration of Large Language
Models.
https://edubruell.github.io/tidyllm/.
Cheng, Joe, Carson Sievert, and Garrick Aden-Buie. 2026.
Shinychat:
Chat UI Component for Shiny.
https://posit-dev.github.io/shinychat/.
Couch, Simon. 2025.
Vitals: Large Language Model Evaluation.
https://vitals.tidyverse.org.
Google Research and Google DeepMind. 2026.
MedGemma 1.5
Technical Report.
https://arxiv.org/abs/2604.05081.
Gruber, Johannes B., and Maximilian Weber. 2025.
Rollama:
Communicate with Ollama to Run Large Language Models Locally.
https://jbgruber.github.io/rollama/.
Inker, Lesley A, Nwamaka D Eneanya, Josef Coresh,
et al. 2021.
“New Creatinine- and
Cystatin C-Based Equations to Estimate GFR without Race.”
N Engl J Med 385 (19): 1737–49.
https://doi.org/10.1056/NEJMoa2102953.
Kalinowski, Tomasz, and Daniel Falbel. 2026.
Ragnar:
Retrieval-Augmented Generation (RAG) Workflows.
https://ragnar.tidyverse.org.
Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group. 2013.
“KDIGO 2012 Clinical Practice Guideline for the
Evaluation and Management of Chronic Kidney Disease.” Kidney
International Supplements 3 (1): 1–150.
https://kdigo.org/wp-content/uploads/2017/02/KDIGO_2012_CKD_GL.pdf.
Sellergren, Andrew, Sahar Kazemzadeh, Tiam
Jaroensri, Atilla Kiraly, Madeleine Traverse, et al. 2025.
MedGemma Technical Report.
https://arxiv.org/abs/2507.05201.
Wickham, Hadley, Joe Cheng, Aaron Jacobs, Garrick Aden-Buie, and Barret
Schloerke. 2026.
Ellmer: Chat with Large Language Models.
https://ellmer.tidyverse.org.
---
title: "Medicine × LLM × R: Showcase"
subtitle: "Demo for the keynote at R Conference Malaysia 2026"
author: "Dr. Wan Nor Arifin<br>Department of Biostatistics & Research Methodology, Universiti Sains Malaysia"
date: "21 November 2026"
output:
  html_document:
    toc: true
    toc_float: true
    toc_depth: 2
    theme: cosmo
    highlight: tango
    code_download: true
    df_print: paged
bibliography: references.bib
link-citations: true
params:
  run_llm: true
---

```{css, echo = FALSE}
h1, h2, h3, h4, .title { color: #165CAA; }
.subtitle { color: #272822; }
a { color: #165CAA; }
.note {
  border-left: 4px solid #165CAA;
  background: #F4F8FC;
  padding: 0.6em 1em;
  margin: 1em 0;
}
.warn {
  border-left: 4px solid #C0392B;
  background: #FDF3F2;
  padding: 0.6em 1em;
  margin: 1em 0;
}
```

```{r setup, include = FALSE}
knitr::opts_chunk$set(echo = TRUE, comment = "#>", message = FALSE, warning = FALSE)
run_llm = isTRUE(params$run_llm)  # FALSE: show code only, without calling any LLM
source("functions.R")
```

# Introduction

This document accompanies the keynote *Medicine × LLM × R* and shows, step by step, how selected medical tasks from the talk can be carried out with large language models (LLMs) in R. The examples progress from basic inference to retrieval-augmented generation (RAG) and a simple application:

| # | Medical task | Package | Example |
|:-:|:-------------|:--------|:--------|
| 1 | Patient communication, translation | `rollama` | Explain laboratory results in simple Malay |
| 2 | Triage / text classification | `rollama` | Zero-shot classification of complaints |
| 3 | Information extraction | `tidyllm` | Structure clinical notes into a data frame |
| 4 | Extraction and evaluation | `ellmer` | Extract variables from notes, evaluate against a gold standard |
| 5 | Decision support (calculation) | `ellmer` | Tool calling: R computes eGFR, the LLM explains |
| 6 | Patient education, grounded in a document | `ragnar` | RAG over a Malay Wikipedia page |
| 7 | Simple application | `ellmer` + `shinychat` + `ragnar` | Patient-education chatbot |

::: {.warn}
All clinical notes are **synthetic**. The code is for demonstration only and is not intended for clinical use.
:::

## Requirements

1. [Ollama](https://ollama.com/download), installed and running, with the following models:

    ```
    ollama pull medgemma1.5:4b   # chat model, Sections 1-4
    ollama pull qwen3.5:4b       # chat model with tool calling, Section 5
    ollama pull bge-m3           # multilingual embedding model, Sections 6-7
    ollama pull hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0   # Malay-capable chat model, Sections 6-7
    ```

2. R packages:

    ```r
    install.packages(c("rollama", "tidyllm", "ellmer", "ragnar", "jsonlite",
                       "shiny", "shinychat", "bslib"))
    ```

    `MJMSTable` is also used for the descriptive table in Section 3.

3. From the talk's repository: the synthetic notes in `data/clinical_notes.csv`, and the helper functions `clean_thought()` and `extract_num()` in `functions.R`.

The LLM outputs below were generated when this document was compiled; they will differ between models, model versions and runs.

```{r model, eval = run_llm}
model = "medgemma1.5:4b"   # chat model used throughout
```

Sections 1–4 use MedGemma, an open-weight medical model [@sellergren2025; @medgemma15].

## R and package versions

```{r version, echo = FALSE}
pkgs = c("rollama", "tidyllm", "ellmer", "ragnar", "jsonlite", "shinychat")
data.frame(
  package = c("R", pkgs),
  version = c(as.character(getRversion()),
              vapply(pkgs, \(p) as.character(packageVersion(p)), character(1))),
  row.names = NULL
)
```

# 1. `rollama`: basic local inference

`rollama` sends prompts to models running locally in Ollama [@rollama]. Nothing leaves the computer, which matters for patient data. A fixed `seed` and `temperature = 0` make the output more reproducible. `clean_thought()` (in `functions.R`) removes the model's reasoning text, keeping only the answer.

```{r rollama-setup, eval = run_llm}
library(rollama)
ping_ollama()        # is Ollama running?
# pull_model(model)    # download once
```

```{r rollama-explain, eval = run_llm}
finding = "HbA1c 8.9%, fasting glucose 11.2 mmol/L, BP 150/95 mmHg."

query(
  paste("Explain these results to a patient in simple Bahasa Melayu,",
        "in 3 sentences. Do not give a diagnosis.", finding),
  model = model,
  output = "data.frame", screen = FALSE,
  model_params = list(seed = 2026, temperature = 0)
) -> res
clean_thought(res$response) |> cat()
```

# 2. `rollama`: zero-shot triage classification

`make_query()` builds a structured prompt from the text, the task and a system role. The result is an ordinary R vector, so the usual tools for diagnostic accuracy and agreement apply.

```{r rollama-triage, eval = run_llm}
complaints = c(
  "Crushing chest pain radiating to the left arm, sweating, 30 minutes.",
  "Mild sore throat for two days, no fever, eating well.",
  "Sudden weakness of the right arm and slurred speech since this morning."
)

q = make_query(
  text   = complaints,
  prompt = "Classify the triage urgency: 'emergency', 'urgent' or 'non-urgent'.",
  system = "You are a triage assistant. Answer with one category only, in lowercase."
)

triage = query(q, model = model, output = "text", screen = FALSE,
               model_params = list(seed = 2026, temperature = 0)) |> clean_thought()

triage_df = data.frame(complaint = complaints, triage = triage)
triage_df
```

Comparison with a reference classification (here, a hypothetical triage nurse):

```{r rollama-triage-table, eval = run_llm}
nurse = c("emergency", "non-urgent", "emergency")
table(LLM = triage_df$triage, Nurse = nurse)
```

# 3. `tidyllm`: structuring clinical notes

`tidyllm` defines the information to extract as a schema; the model is constrained to return JSON that follows it [@tidyllm; @arifin2026workshop].

```{r tidyllm-schema, eval = run_llm}
library(tidyllm)
library(jsonlite)

patient_schema = tidyllm_schema(
  name = "PatientHistory",
  patient_name = field_chr("Patient's name; empty if not stated"),
  age = field_dbl("Age in years. If given in months, convert to a fraction of a year."),
  gender = field_fct("Gender of the patient", .levels = c("male", "female")),
  symptoms = field_chr("Current symptoms reported by the patient", .vector = TRUE)
)
```

One note:

```{r tidyllm-one, eval = run_llm}
raw_text = "Mrs. Amy, a 35-year-old lady came to the outpatient clinic with the
            complaint of fever for three days and cough."

result = llm_message(paste("Extract the patient's note according to the schema.", raw_text)) |>
  chat(.provider = ollama(), .model = model, .json_schema = patient_schema,
       .temperature = 0, .seed = 2026)
# Cloud alternative: chat(.provider = gemini(), .model = "gemini-3.1-flash-lite", ...)

get_reply(result) |> minify()
```

Several notes, saved as JSON Lines and then read into a data frame:

```{r tidyllm-many, eval = run_llm}
notes = read.csv("data/clinical_notes.csv")
output_file = "pt_data.jsonl"
if (file.exists(output_file)) file.remove(output_file)   # start a new file on each run

for (note in notes$note) {
  tryCatch({
    llm_message(paste("Extract the patient's note according to the schema.", note)) |>
      chat(.provider = ollama(), .model = model, .json_schema = patient_schema,
           .temperature = 0, .seed = 2026) |>
      get_reply() |> minify() |>
      cat("\n", file = output_file, append = TRUE, sep = "")
  }, error = function(e) cat("Error:", conditionMessage(e), "\n"))
}

patient_data = stream_in(file(output_file), verbose = FALSE)   # read the JSON Lines file
patient_data
```

From unstructured text to the descriptive table we already know. 

::: {.note}
You may download MJMSTable R package from: <https://github.com/wnarifin/MJMSTable>.
:::

```{r tidyllm-table, eval = run_llm}
MJMSTable::descriptive_tbl(patient_data, included_var = c("age", "gender"))
```

# 4. `ellmer`: extraction and evaluation

`ellmer` provides one interface to many providers, cloud and local [@ellmer]. The schema is written with R types, and `parallel_chat_structured()` returns one row per note. For the numeric variables, the model copies the relevant text from the note (e.g. `'132 umol/L'`), or an empty string if the value is not reported.

The synthetic notes contain deliberate traps: an ex-smoker, a note without smoking status, missing laboratory results and "percent" written in words.

```{r notes}
notes = read.csv("data/clinical_notes.csv")
notes[, c("id", "note")]
```

```{r ellmer-extract, eval = run_llm}
library(ellmer)

type_patient = type_object(
  age = type_number("Age in years, as written in the note"),
  sex = type_enum(values = c("male", "female"),
                  description = "Sex of the patient, from words such as man, lady, Encik, Puan"),
  bp_text = type_string("Blood pressure value and unit only, copied exactly from the note, e.g. '150/95 mmHg'. Empty string if not reported."),
  hba1c_text = type_string("HbA1c value and unit only, copied exactly from the note, e.g. '8.9%'. Empty string if not reported."),
  creatinine_text = type_string("Serum creatinine value and unit only, copied exactly from the note, e.g. '132 umol/L'. Empty string if not reported."),
  smoker = type_enum(values = c("yes", "no", "unknown"),
                     description = "Is the patient a CURRENT smoker? Ex-smokers and never-smokers are 'no'. 'unknown' if smoking is not mentioned.")
)

chat = chat_ollama(
  model = model,
  system_prompt = "You extract data from clinical notes. Do not guess values that are not stated.",
  params = params(temperature = 0, seed = 2026)
)

extracted = parallel_chat_structured(chat, as.list(notes$note), type = type_patient)
extracted
```

A small model often returns wrong or missing numbers when asked for them directly, and may fill in typical values that are not in the note. The LLM is therefore used only to locate the text, and R extracts the numbers from it with `extract_num()` (in `functions.R`). Text without a value, e.g. "No HbA1c done this visit", gives `NA`.

```{r extract-numbers, eval = run_llm}
extracted$sbp        = extract_num(extracted$bp_text, 1)
extracted$dbp        = extract_num(extracted$bp_text, 2)
extracted$hba1c      = extract_num(extracted$hba1c_text)
extracted$creatinine = extract_num(extracted$creatinine_text)
extracted
```

## Evaluation against the gold standard

The gold standard is stored alongside the notes. Agreement is defined as equal values, or missing in both.

```{r gold}
gold = notes[, c("age", "sex", "sbp", "dbp", "hba1c", "creatinine", "smoker")]
agree = function(x, y) (is.na(x) & is.na(y)) | (!is.na(x) & !is.na(y) & x == y)
```

Field-level accuracy (%):

```{r eval-field, eval = run_llm}
acc = sapply(names(gold), function(x) mean(agree(extracted[[x]], gold[[x]])))
round(acc * 100, 2)
```

For systematic evaluation across tasks, models and repeated runs, see `vitals` [@vitals].

# 5. `ellmer`: tool calling, let R do the maths

LLMs are unreliable calculators. An R function is exact and can be validated; the LLM decides **when** to call it and explains the result. The function below implements the race-free CKD-EPI 2021 creatinine equation [@inker2021].

```{r egfr}
egfr_ckdepi2021 = function(creatinine_umol, age, sex) {
  scr = creatinine_umol / 88.4                 # umol/L -> mg/dL
  female = sex == "female"
  k = ifelse(female, 0.7, 0.9)
  a = ifelse(female, -0.241, -0.302)
  142 * pmin(scr / k, 1)^a * pmax(scr / k, 1)^-1.200 *
    0.9938^age * ifelse(female, 1.012, 1)
}

egfr_ckdepi2021(132, age = 64, sex = "female") |> round(1)
```

```{r egfr-tool, eval = run_llm}
tool_egfr = tool(
  egfr_ckdepi2021,
  name = "egfr_ckdepi2021",
  description = "Estimate GFR (mL/min/1.73 m2) using the race-free CKD-EPI 2021 creatinine equation.",
  arguments = list(
    creatinine_umol = type_number("Serum creatinine in umol/L"),
    age = type_number("Age in years"),
    sex = type_enum(values = c("male", "female"), description = "Sex of the patient")
  )
)
```

Here we use qwen3.5:4b, because medgemma1.5 does not support tool calling. The answer should give the KDIGO CKD GFR category [@kdigo2012]: an eGFR of 38.9 mL/min/1.73 m² is category G3b, moderately to severely decreased.

```{r use-egfr-tool, eval = run_llm}
model_tool = "qwen3.5:4b"
# temperature > 0 for a thinking model: at temperature = 0 it can get stuck in a
# thinking loop; the fixed seed keeps the run reproducible
chat = chat_ollama(model = model_tool, params = params(temperature = 0.5, seed = 2026))
chat$register_tool(tool_egfr)

t0 = Sys.time()
res_tool = chat$chat("Puan Aminah is a 64-year-old woman with serum creatinine 132 umol/L.
                     What is her eGFR, which KDIGO CKD GFR category and
                     which clinical term is it?", echo = "none")
time_tool = as.numeric(difftime(Sys.time(), t0, units = "secs"))

res_tool
cat("Time taken (with tool):", round(time_tool, 1), "seconds\n")
```

For comparison, the same question without the tool. If you try this out, you will notice that qwen3.5:4b takes considerably longer, as it reasons through the calculation itself.

```{r egfr-notool, eval = run_llm}
chat_no_tool = chat_ollama(model = model_tool, params = params(temperature = 0.5, seed = 2026))

t0 = Sys.time()
res_no_tool = chat_no_tool$chat("Puan Aminah is a 64-year-old woman with serum creatinine 132 umol/L.
                                Calculate her eGFR with CKD-EPI 2021 creatinine equation.
                                Which KDIGO CKD GFR category and which clinical term is it?",
                                echo = "none")
time_no_tool = as.numeric(difftime(Sys.time(), t0, units = "secs"))

res_no_tool
cat("Time taken (without tool):", round(time_no_tool, 1), "seconds\n")
```

Time and tokens used, with and without the tool:

```{r egfr-compare, eval = run_llm}
tok_tool = chat$get_tokens()
tok_no_tool = chat_no_tool$get_tokens()

data.frame(
  run = c("With tool", "Without tool"),
  seconds = round(c(time_tool, time_no_tool), 1),
  assistant_turns = c(nrow(tok_tool), nrow(tok_no_tool)),
  input_tokens = c(sum(tok_tool$input, na.rm = TRUE), sum(tok_no_tool$input, na.rm = TRUE)),
  output_tokens = c(sum(tok_tool$output, na.rm = TRUE), sum(tok_no_tool$output, na.rm = TRUE))
)
```

# 6. `ragnar`: grounding answers in a document (RAG)

Retrieval-augmented generation (RAG) first retrieves relevant passages from a document store, then lets the LLM answer from them. This reduces hallucination and allows the answer to cite its source [@ragnar]. Here the store is built from the Malay Wikipedia page *Kencing manis*.

::: {.note}
`read_as_markdown()` cannot read the Wikipedia URL directly, because Wikipedia rejects the default User-Agent of the Python library it uses (HTTP 403). Instead, the article is downloaded as plain text through the Wikipedia API with `download.file()`, which also leaves out the page menus and sidebars, and saved as a markdown file for `read_as_markdown()`.
:::

```{r ragnar-download, eval = run_llm}
library(ragnar)

# Plain-text article from the Wikipedia API (no menus or sidebars)
url = "https://ms.wikipedia.org/w/api.php?action=query&prop=extracts&explaintext=1&redirects=1&format=json&titles=Kencing_manis"
json_file = "kencing_manis.json"
md_file = "kencing_manis.md"
download.file(url, destfile = json_file, mode = "wb")
writeLines(jsonlite::fromJSON(json_file)$query$pages[[1]]$extract,
           md_file, useBytes = TRUE)
```

The store is a DuckDB file. It should be on a local disk: cloud-synced folders (e.g. pCloud Drive) can break DuckDB's temporary write-ahead (`.wal`) file.

```{r ragnar-store, eval = run_llm}
store_location = path.expand("~/ragnar_store/kencing_manis.duckdb")   # local disk
dir.create(dirname(store_location), showWarnings = FALSE)

store = ragnar_store_create(
  store_location,
  embed = \(x) embed_ollama(x, model = "bge-m3"),   # multilingual embeddings
  overwrite = TRUE                                  # rebuild from scratch
)

chunks = read_as_markdown(md_file) |> markdown_chunk()
chunks

ragnar_store_insert(store, chunks)
ragnar_store_build_index(store)
```

Retrieval of the most relevant chunks for a question:

```{r ragnar-retrieve, eval = run_llm}
ragnar_retrieve(store, "Apakah tanda-tanda kencing manis?", top_k = 3)
```

The store is registered as a tool, so the LLM searches it before answering. Here we use [Apertus-SEA-LION-v4-8B](https://huggingface.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF) from AI Singapore, a model that supports Bahasa Melayu:

```{r ragnar-chat, eval = run_llm}
model_my = "hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0"
chat = chat_ollama(
  model = model_my,
  system_prompt = paste(
    "You explain diabetes to patients in simple Malay or English.",
    "Search the knowledge store before answering and cite it.",
    "You do not diagnose. Always advise seeing a doctor."
  ),
  params = params(temperature = 0, seed = 2026)
)
ragnar_register_tool_retrieve(chat, store, top_k = 5)
chat$chat("Apakah tanda-tanda awal kencing manis?")
```

The same steps apply to other documents, for example a local clinical practice guideline in PDF: `read_as_markdown("path/to/cpg.pdf")`.

# 7. `ellmer` + `shinychat` + `ragnar`: a patient-education app

A Shiny app cannot run inside a static HTML document. Run the code below in R (or `code/06_ellmer_shinychat_ragnar_app.R`) after building the store in Section 6 [@shinychat].

```{r app, eval = FALSE}
library(shiny)
library(shinychat)
library(ellmer)
library(ragnar)

store_location = path.expand("~/ragnar_store/kencing_manis.duckdb")   # as in Section 6
store = ragnar_store_connect(store_location, read_only = TRUE)

ui = bslib::page_fillable(
  bslib::card(
    bslib::card_header("Pembantu pendidikan pesakit: kencing manis (demo)"),
    chat_ui("chat")
  )
)

model_my = "hf.co/aisingapore/Apertus-SEA-LION-v4-8B-IT-GGUF:Q8_0"
server = function(input, output, session) {
  chat = chat_ollama(
    model = model_my,
    system_prompt = "You explain diabetes to patients in simple Malay or English.
      Search the knowledge store before answering and cite it.
      You do not diagnose. Always advise seeing a doctor."
  )
  ragnar_register_tool_retrieve(chat, store, top_k = 5)
  chat_server("chat", chat, history = FALSE)
}

shinyApp(ui, server)
```

For a quick personal chat in the browser, without writing an app:

```{r live-browser, eval = FALSE}
chat = chat_ollama(model = model_my)
ragnar_register_tool_retrieve(chat, store)
live_browser(chat)
```

::: {.warn}
**Not a virtual doctor.** Keep a human in the loop and validate any tool before clinical use. Wikipedia is used here only as an openly licensed example; for practice, use authoritative sources such as clinical practice guidelines.
:::

# References
