Sources and licences
What an answer can draw on, in order of authority. Sources are retrieved for each question the model answers; exam questions with options are answered without labels, the guideline library or literature. The references under an answer list what was retrieved and link to the original where there is one; Ayulex's internal summaries are its own notes and have none. Each source below is used under its published terms.
US FDA prescribing information
DailyMed, through openFDA
- Used for
- Doses, renal and hepatic adjustment, contraindications, boxed warnings and labelled interactions. These are US labels; availability and dosing can differ elsewhere.
- Licence
- US government work, not subject to copyright in the US.
- Kept current
- Read from Ayulex's copy of openFDA's labels, most of them reloaded each month, or from openFDA directly for a medicine not in the copy: for medicines a question names, before the answer is written, and for those it recommends, after. Not read for exam questions with options. The answer links to the label and shows its effective date.
FDA Drug Safety Communications
US Food and Drug Administration
- Used for
- New warnings, restrictions and withdrawals, shown under an answer that recommends a drug or class they concern.
- Licence
- US government work, not subject to copyright in the US.
- Kept current
- 306 communications since January 2010, with what each one says, taken from the FDA's lists and archived copies on 2 November 2026. The FDA's MedWatch and "What's New: Drugs" feeds are read once a day for newer ones.
MHRA Drug Safety Update
Medicines and Healthcare products Regulatory Agency, United Kingdom
- Used for
- UK safety advice on medicines, shown first under an answer for a doctor practising in the UK, and alongside FDA notices outside the US.
- Licence
- Open Government Licence v3.0. Contains public sector information licensed under that licence.
- Kept current
- 870 articles, taken from GOV.UK on 2 November 2026.
Drugs withdrawn from the US market
21 CFR 216.24, US Code of Federal Regulations
- Used for
- Flags a recommended drug for a doctor practising in the US when its entry covers every product containing it. Entries limited to a form or strength are shown when the drug is looked up.
- Licence
- US government work, not subject to copyright in the US.
- Kept current
- 81 entries, as of 27 October 2026.
Prohibited and restricted drugs
Central Drugs Standard Control Organisation, Government of India
- Used for
- Combinations and single drugs prohibited under Section 26A of the Drugs and Cosmetics Act, with the status CDSCO gives each: prohibited, restricted, under examination, stayed or quashed.
- Licence
- Notifications published in the Gazette of India, which Section 52(1)(q) of the Copyright Act, 1957 permits reproducing.
- Kept current
- 723 entries from CDSCO's published documents, to the notification of 18 August 2026. Court orders after a document's date are not reflected, and answers say so.
National List of Essential Medicines 2022
Ministry of Health and Family Welfare, Government of India
- Used for
- Whether a medicine is essential in India, at which level of care (primary, secondary, tertiary) and in which forms and strengths.
- Licence
- Schedule I of the Drugs (Prices Control) Order, 2013, published in the Gazette; reproducible under Section 52(1)(q).
- Kept current
- The 2022 list, the latest published (13 September 2022): 383 medicines.
NPPA Pharma Sahi Daam
National Pharmaceutical Pricing Authority, Government of India
- Used for
- Indian brand names: recognising them in questions as the medicines they contain, and each product's composition, pack, company and MRP, as NPPA lists them, when a brand is asked about.
- Licence
- Read from NPPA's public price search, slowly and identified, and reproduced under NPPA's copyright policy, which allows its material to be reproduced accurately, in no misleading context, with the source prominently acknowledged.
- Kept current
- 95,351 products of 63,203 brands from 827 companies, with MRPs as listed for Aug 2026, read on 5 November 2026.
NPPA revised MRPs of anti-cancer drugs
National Pharmaceutical Pricing Authority, Government of India
- Used for
- Indian brand names of anti-cancer drugs: recognising them in questions as the drug they contain, and each brand's company, strength, pack and MRP as the list gives them.
- Licence
- Reproduced under NPPA's copyright policy, which allows its material to be reproduced accurately, in no misleading context, with the source prominently acknowledged.
- Kept current
- 1,136 products, 671 brand names of 120 drugs, effective 9 November 2026, based on data the companies sent NPPA up to 4 November 2026. Every row was read again from the PDF and checked. Brands of other medicines are not included.
LactMed
US National Library of Medicine
- Used for
- What is known about each medicine in breastfeeding: levels in milk, effects in breastfed infants, alternatives.
- Licence
- Public domain under NLM's terms, with attribution to the authoring agency. LactMed is a registered trademark of the US Department of Health and Human Services.
- Kept current
- 1,955 medicines, built from NLM's archive on 17 October 2026.
LiverTox
US National Institute of Diabetes and Digestive and Kidney Diseases
- Used for
- How likely a medicine is to injure the liver (likelihood categories A to E), and the usual pattern of injury.
- Licence
- Public domain under NLM's terms, with attribution to the authoring agency.
- Kept current
- 1,108 medicines, built from NLM's archive on 17 October 2026.
Guideline library
Ayulex, from each publisher's own documents
- Used for
- Passages from clinical guidelines, national programme documents and review articles, found by searching with the question and given to the answering model as numbered sources; review articles are labelled as such. Not searched for exam questions with options.
- Licence
- Only documents whose terms allow them to be stored and quoted: Indian programme documents (NTEP, NCVBDC, NCDC, NHM, ICMR), reproduced with the source acknowledged as each website's copyright policy allows; US government works in the public domain (CDC, NCCIH and other NIH institutes); NCI's PDQ summaries, whose text can be used freely; GOV.UK guidance under the Open Government Licence v3.0; and guidelines and reviews from Europe PMC under CC BY or CC0, or in the public domain. The licence is stored with every passage.
- Kept current
- Reloaded from each publisher on the 2nd of each month.
PubMed
US National Library of Medicine
- Used for
- Guidelines, systematic reviews, meta-analyses and trials, preferred in that order. Other article types are used only when few of those match; each reference shows its article type.
- Licence
- Records through NCBI's E-utilities. Abstracts can be under publisher copyright, so only short excerpts are shown, with a link to the record.
- Kept current
- Searched when a question asks for guidelines, trials, studies or recent evidence; for a question in another language or with an attachment, when the planning step asks for it; and by the answering model in the medication review and SOAP note presets. Not searched for exam questions with options. Citations carry real PubMed identifiers.
RxNorm drug names
US National Library of Medicine
- Used for
- Recognising drug names and US brand names in questions and answers, so the checks apply when a medicine is written by brand (an allergy to Augmentin is an allergy to amoxicillin and clavulanate).
- Licence
- RxNorm Current Prescribable Content, which NLM provides without licensing restrictions. This product uses publicly available data courtesy of the U.S. National Library of Medicine (NLM); NLM is not responsible for the product and does not endorse or recommend it.
- Kept current
- 3,316 generic names and 3,332 US brand names, from the release of 5 October 2026. Indian brand names come from NPPA's Pharma Sahi Daam and its list of anti-cancer drugs (above). Generic names on India's own lists that RxNorm lacks (NLEM 2022, CDSCO's prohibited drugs, NPPA's price lists and its retail prices for new drugs in 2014 and since 2019) are added by hand, each one checked.
Orphanet rare diseases
Orphanet (INSERM)
- Used for
- For a case recognised as hard, ranking the rare diseases whose clinical features best match the case's findings, read from the question and from laboratory values by reference range. Each match lists the typical features the case does not mention, as what to check next. A prompt to consider, not a diagnosis.
- Licence
- Orphadata: Free access data from Orphanet. © INSERM 1999. Available on http://www.orphadata.com. Licensed under CC BY 4.0.
- Kept current
- 4,355 rare diseases with 8,758 annotated clinical features, from the data of 23 June 2026 (version 1.3.42 / 4.1.8 [2025-03-03]).
ClinicalTrials.gov
US National Library of Medicine
- Used for
- For questions about trials or about new, investigational or latest treatments: registered trials with their design, status, size and whether results are posted, so the answer can describe what is being studied now. A registry entry is not cited as evidence of benefit unless its results are posted.
- Licence
- Public domain (US government work), through the ClinicalTrials.gov API.
- Kept current
- Searched for a question that mentions trials or asks for the latest evidence, and for one about new or investigational treatments that names a medicine; registry entries are kept for a day.
Clinical calculators
Ayulex, from each score's original publication
- Used for
- Scores, clearances and conversions, computed in code from the published formula with every input shown. The answering model is told to use them rather than work out a score itself; a score shown without a calculator was not computed in code.
- Licence
- Ayulex code. The publication behind each formula is listed below.
- Kept current
- 73 calculators, each tested against worked examples on every change.
Reviewed corrections
Users of Ayulex, confirmed by a reviewer
- Used for
- A correction a doctor made to an earlier answer, after a reviewer checked it against the source it cites. Given to the answering model for later questions about the same medicines, as a numbered source.
- Licence
- Statements submitted by users for this purpose; each rests on the published source it cites.
- Kept current
- Added when a reviewer approves one. Nothing pending or rejected is used.
Ayulex internal summaries
Ayulex
- Used for
- Short notes written from published guidance, used as a starting point for common emergencies and high-alert drugs. Labelled as internal wherever they appear. They contain doses, which the answering model may use; confirm any dose that rests on them against the label.
- Licence
- Written by Ayulex.
- Kept current
- 36 protocol summaries and 20 drug safety summaries. Not yet reviewed by an independent clinician.
Not used: material whose licence or published terms do not allow storing and quoting it, such as commercial brand databases, Jan Aushadhi product lists and commercial references; these will be added only under a licence or written permission. ICMR, NHM, NTEP and NCDC documents are used under each publisher's stated reuse terms, and NPPA's price lists under NPPA's.
Checks on every answer
After an answer is written, up to six medicines it recommends are checked in code against the patient details in the conversation and against the sources above; labels, safety notices, LactMed and LiverTox are read for the first four. What does not match is shown under the answer with the source sentence, so you can judge it yourself.
| Check | What is looked for |
|---|---|
| Doses | Doses written as an amount or an amount per kg are looked for in the material read for the answer, your own messages included; how often a dose is given is not compared, and a dose written as a rate with a slash (such as mg/day or units/h, other than per kg) or with its unit spelt out (such as micrograms) is not read. Up to eight doses found in none of it are listed. |
| Citations | Up to 12 cited sentences per answer, those with numbers first, are compared with the text of their sources in a separate step, which sees only those sentences and their sources. |
| Allergies | A recommended drug, or a drug of the same class, that the case says the patient is allergic to. |
| Kidney function | The label's renal dosing or contraindication at the patient's creatinine clearance, with the sentence quoted. |
| Pregnancy and breastfeeding | The label's pregnancy warnings when the case says the patient is pregnant, and LactMed's summary when it involves breastfeeding. |
| Liver | LiverTox's likelihood of liver injury when the case involves liver disease. |
| Interactions | Labelled interactions between recommended drugs and the patient's current medicines. |
| Boxed warnings | The label's boxed warning, quoted, for each recommended drug that carries one. |
| FDA safety communications | Warnings, restrictions or withdrawals the FDA has issued since January 2010 for the drug or its class. |
| India's prohibited list | Whether the exact combination, strength and form is prohibited or restricted by CDSCO, with the notification and its current status. |
| Dose for this patient | A mg/kg dose multiplied out for the patient's weight, with the daily total, compared with the label's maximum; any amount the answer states is checked against the multiplication, and a dose 1,000 times off the label (mg for mcg) is flagged. |
| Can't-miss diagnoses | For a patient presenting with chest pain, headache, abdominal pain, breathlessness, fever and other complaints, whether the answer addresses each time-critical cause on Ayulex's checklist. |
Before and after the answer
The case in one line
For a patient case, the details Ayulex reads from the conversation in code (age, sex, weight, kidney function, allergies, regular medicines) are shown above the answer so a misreading can be corrected, and the answering model is given them. It costs no model call.
Details that decide the medicine
When a medicine or a dose is asked for a patient, code checks whether the details that decide it were given: age, pregnancy or breastfeeding, kidney function, allergies and other medicines. Any that are missing are named to the answering model, which is told not to assume them, to say which are missing, and to answer for each case where the choice differs, or to say there is not enough information.
Repair, not only flag
If the safety check finds a serious problem the answer does not address, the sentences that recommend the drug are rewritten in a separate step, using only the finding and the label sentence. Code then rejects any rewrite that adds a dose, a drug or a citation that is in neither. The revised sentence replaces the original on screen, marked "Revised", and both are shown below the answer with the reason.
How well the answer is supported
The checks are summed up in one line: well, partly or weakly supported, with the reasons, such as a cited statement its source does not support, a dose found in no retrieved source, a serious finding left open, or details not given. An answer is weak when a statement its source does not support contains a dose or number, when two or more doses appear in no retrieved source, or when a serious finding is left open. A cited statement without a number that its source does not support usually has its citation removed, and makes the answer partly supported. The citation check compares facts with their sources, not the judgement around them, such as calling a drug preferred. It is a summary of the checks, not a probability.
Learning from corrections
A signed-in doctor who marks an answer wrong can say what is correct and give a source: a PubMed ID, a DOI or a link. It waits for review. A reviewer checks it against the source and approves or rejects it; approval records who reviewed it and the medicines it names, and a statement that names no medicine cannot be approved. From then on, every question about one of those medicines retrieves it, and the answering model is given it as a numbered source, listed as a reviewed correction. Nothing pending or rejected is ever used, and no model is retrained: the correction is part of Ayulex's own library.
Calculators and their published sources
Each of the 73 calculators is written once, from the publication listed, and that one definition is used by the chat, the calculator drawer and this website. Each is tested against worked examples on every change; the results are on the evaluation page.
| Calculator | Published source | Tests |
|---|---|---|
| Wells' Criteria for Pulmonary Embolism | Wells PS et al. Thromb Haemost 2000;83:416-420. Two-level validation: van Belle A et al. JAMA 2006;295:172-179. | 2 |
| Wells' Criteria for DVT | Wells PS et al. N Engl J Med 2003;349:1227-1235. | 2 |
| CHA₂DS₂-VASc Score | Lip GYH et al. Chest 2010;137:263-272. | 4 |
| HEART Score | Six AJ, Backus BE, Kelder JC. Neth Heart J 2008;16:191-196. Validation: Backus BE et al. Int J Cardiol 2013;168:2153-2158. | 2 |
| TIMI Risk Score (UA/NSTEMI) | Antman EM et al. JAMA 2000;284:835-842. | 2 |
| HAS-BLED Bleeding Risk | Pisters R et al. Chest 2010;138:1093-1100. | 2 |
| Corrected QT Interval (QTc) | Bazett HC. Heart 1920;7:353-370. Fridericia LS. Acta Med Scand 1920;53:469-486. Thresholds: Rautaharju PM et al. Circulation 2009;119:e241-e250. | 2 |
| Cockcroft-Gault Creatinine Clearance | Cockcroft DW, Gault MH. Nephron 1976;16:31-41. | 3 |
| eGFR — CKD-EPI 2021 (creatinine) | Inker LA et al. N Engl J Med 2021;385:1737-1749. | 3 |
| Anion Gap, Albumin Correction and Delta Ratio | Figge J et al. Crit Care Med 1998;26:1807-1810 (albumin correction). | 3 |
| Fractional Excretion of Sodium (FENa) | Espinel CH. JAMA 1976;236:579-581. | 2 |
| Sodium Corrected for Hyperglycaemia | Katz MA. N Engl J Med 1973;289:843-844. Hillier TA et al. Am J Med 1999;106:399-403. | 2 |
| Calcium Corrected for Albumin | Payne RB et al. Br Med J 1973;4:643-646. | 2 |
| CURB-65 Pneumonia Severity | Lim WS et al. Thorax 2003;58:377-382. | 2 |
| Glasgow Coma Scale | Teasdale G, Jennett B. Lancet 1974;2:81-84. | 3 |
| qSOFA | Seymour CW et al. JAMA 2016;315:762-774. | 2 |
| Mean Arterial Pressure | Standard formula: diastolic + (systolic − diastolic) ÷ 3. Target in septic shock: Evans L et al. Intensive Care Med 2021;47:1181-1247. | 2 |
| Parkland Formula | Baxter CR, Shires T. Ann N Y Acad Sci 1968;150:874-894. | 4 |
| MELD-Na (2016) | Kim WR et al. N Engl J Med 2008;359:1018-1026. Mortality bands: Wiesner R et al. Gastroenterology 2003;124:91-96. | 3 |
| Child-Pugh Score | Pugh RN et al. Br J Surg 1973;60:646-649. | 2 |
| Glasgow-Blatchford Bleeding Score | Blatchford O, Murray WR, Blatchford M. Lancet 2000;356:1318-1321. | 4 |
| Alvarado Score | Alvarado A. Ann Emerg Med 1986;15:557-564. | 2 |
| Modified Centor (McIsaac) Score | McIsaac WJ et al. CMAJ 1998;158:75-83; JAMA 2004;291:1587-1595. | 2 |
| Maintenance Fluids (Holliday-Segar 4-2-1) | Holliday MA, Segar WE. Pediatrics 1957;19:823-832. | 3 |
| Apgar Score | Apgar V. Curr Res Anesth Analg 1953;32:260-267. AAP/ACOG Committee Opinion 644, 2015. | 2 |
| Caprini Risk Assessment Model for VTE | Caprini JA. Dis Mon 2005;51:70-78; items and risk categories as in Gould MK et al. Chest 2012;141(2 Suppl):e227S-e277S (ACCP 9th edition, Tables 5 and 7). | 4 |
| Revised Cardiac Risk Index (Lee) | Lee TH et al. Circulation 1999;100:1043-1049. Predictors, and risk by class of myocardial infarction, cardiac arrest or cardiac death in the original cohort (re-estimated by Devereaux et al.), as given in Roshanov PS et al. BMJ Open 2017;7:e013510. | 3 |
| Simplified Pulmonary Embolism Severity Index (sPESI) | Jiménez D et al. Arch Intern Med 2010;170:1383-1389. Items as listed in Palas M et al. TH Open 2022;6:e347-e353; risk classification as in the Brazilian Thoracic Association's guideline (Amado VM et al. J Bras Pneumol 2025;51:e20240314). | 2 |
| PERC Rule for Pulmonary Embolism | Kline JA et al. J Thromb Haemost 2004;2:1247-1255 and 2008;6:772-780. Criteria as listed in Freund Y et al. Trials 2015;16:537 and Streiff MB et al. J Thromb Thrombolysis 2016;41:32-67. | 2 |
| Bedside Schwartz eGFR for Children | Schwartz GJ et al. J Am Soc Nephrol 2009;20:629-637. Equation (0.413 × height in cm ÷ creatinine in mg/dL) as given in Batte A et al. BMC Nephrol 2020;21:417. | 2 |
| NEWS2 (National Early Warning Score 2) | Royal College of Physicians. National Early Warning Score (NEWS) 2, 2017. Chart and responses as tabulated in Zhang K et al. Front Med 2021;8:699880 (Table 1) and Martín-Rodríguez F et al. J Clin Med 2020;9:1156 (Table 1). | 4 |
| ABCD² Score for TIA | Johnston SC et al. Lancet 2007;369:283-292. Points as tabulated in Ildstad F et al. Stroke Res Treat 2021;2021:8845898 (Table 1); risk strata as in Galvin R et al. BMC Res Notes 2014;7:281; high-risk TIA (4 or more) as in the European Stroke Organisation's expedited recommendations (2021). | 3 |
| SOFA Score (Sequential Organ Failure Assessment) | Vincent JL et al. Intensive Care Med 1996;22:707-710. Criteria as tabulated in Lambden S et al. Crit Care 2019;23:374 (Table 1); sepsis definition as in the Japanese Clinical Practice Guidelines for Management of Sepsis and Septic Shock 2020 (CQ1-2). | 4 |
| FIB-4 Index for Liver Fibrosis | Sterling RK et al. Hepatology 2006;43:1317-1325. Formula as given in Li Q et al. Oncotarget 2017;8:45768-45776; cut-offs 1.30 and 2.67 as in Gabriel-Medina P et al. Intern Emerg Med 2024;19:745-756; stepwise use as in the EASL-EASD-EASO MASLD guideline (2024). | 3 |
| Maddrey Discriminant Function for Alcoholic Hepatitis | Maddrey WC et al. Gastroenterology 1978;75:193-199; Carithers RL et al. Ann Intern Med 1989;110:685-690. Formula and threshold as in Rachakonda V et al. F1000Res 2020;9:97 (Table 1). | 3 |
| BISAP Score for Acute Pancreatitis | Wu BU et al. Gut 2008;57:1698-1703. Criteria and accuracy as in Gao W et al. PLoS One 2015;10:e0130412. | 2 |
| Padua Prediction Score for VTE in Medical Inpatients | Barbar S et al. J Thromb Haemost 2010;8:2450-2457. Items, points and the Padua study's VTE rates as given in Lavon O et al. Sci Rep 2022;12:6121. | 2 |
| 4Ts Score for Heparin-Induced Thrombocytopenia | Lo GK et al. J Thromb Haemost 2006;4:759-765. Categories and probabilities as tabulated in Fathi M. Glob Cardiol Sci Pract 2018;2018:15 (Table 1) and Patriarcheas V et al. Cureus 2020;12:e7385 (Table 2). | 3 |
| BMI, Body Surface Area and Ideal Body Weight | BMI categories: WHO; Asian thresholds 23 and 27.5 from the WHO expert consultation (2004), as tabulated in Murphy M et al. Curr Obes Rep 2017;6:314-323. Body surface area: Mosteller RD. N Engl J Med 1987;317:1098, as given in Dobinson HC et al. Clin Infect Dis 2017;64:1066-1073. Ideal body weight: Devine BJ. Drug Intell Clin Pharm 1974;8:650-655, as given in Min KL et al. PLoS One 2018;13:e0207588. | 3 |
| PHQ-9 Depression Questionnaire | Kroenke K et al. J Gen Intern Med 2001;16:606-613. Severity bands as in Choi EPH et al. Int J Environ Res Public Health 2020;17:3740; cut-off of 10 as in Muñoz-Navarro R et al. BMC Psychiatry 2017;17:291. The PHQ measures may be reproduced without permission (AlHadi AN et al. Ann Gen Psychiatry 2017;16:32). | 3 |
| GAD-7 Anxiety Questionnaire | Spitzer RL et al. Arch Intern Med 2006;166:1092-1097. Items and severity bands as in Sun J et al. Healthcare (Basel) 2021;9:1709. | 2 |
| NIH Stroke Scale (NIHSS) | Brott T et al. Stroke 1989;20:864-870; the scale is published by the US National Institute of Neurological Disorders and Stroke. Items as tabulated in Budinčević H et al. Medicina 2022;58:1541 (Table 2) and Hasan TF et al. Biomedicines 2021;9:1486 (Table 1); severity bands as in Kogan E et al. BMC Med Inform Decis Mak 2020;20:8. | 6 |
| APACHE II Score | Knaus WA et al. Crit Care Med 1985;13:818-829. Points as tabulated in Thakur R et al. Cureus 2023;15 (Table 7) and Zali AR et al. Glob J Health Sci 2012;4:179-184 (Table 1), both CC BY. | 4 |
| TB First-Line Doses by Weight Band (NTEP) | Adults: NTEP, Training Module on Extrapulmonary Tuberculosis (2023), weight band table. Children: NTEP and NACO, Guidelines on Prevention and Management of TB in PLHIV at ART Centres (2016), Table 6. | 8 |
| TB Preventive Treatment Doses (NTEP) | NTEP, Guidelines for Programmatic Management of TB Preventive Treatment in India (2021), Tables 5.1 and 8.1. | 10 |
| Severe Dehydration, Plan C (IMNCI) | MoHFW, IMNCI Chart Booklet for Medical Officers (2023): Plan C, treat severe dehydration quickly. | 2 |
| Antibiotics for Severe Acute Malnutrition (F-IMNCI) | MoHFW, F-IMNCI Chart Booklet (2023), Chart 9.6: recommended antibiotics for children with SAM. | 3 |
| Status Epilepticus Drug Doses in Children | Convulsive status epilepticus in children: adapted clinical practice guideline (2017, CC BY; PMC5726823), table of drug doses. | 7 |
| Convulsions in a Young Infant (IMNCI) | MoHFW, IMNCI Chart Booklet for Medical Officers (2023): treat convulsion in the young infant. | 3 |
| Dextrose and Fluid Bolus for a Sick Child (F-IMNCI) | MoHFW, F-IMNCI Chart Booklet (2023): emergency management of the sick child. | 2 |
| Severe Malaria, Parenteral Doses (NVBDCP) | NVBDCP, National Drug Policy on Malaria (2013): treatment of severe malaria. | 4 |
| Artemether–Lumefantrine (ACT-AL) by Weight | US FDA prescribing information, artemether and lumefantrine tablets (DailyMed, March 2026), sections 2.2 and 2.3. | 4 |
| Primaquine for P. vivax (NCVBDC) | NCVBDC, Malaria Training Module for Multi-Purpose Health Workers (2025), Table 1. | 2 |
| Rabies Immunoglobulin, Maximum Dose (NCDC) | NCDC, Training Module on Rabies for Medical Officers (2023): passive immunization with rabies immunoglobulin. | 2 |
| Adrenaline for Anaphylaxis by Age (MoHFW) | MoHFW, Operational Guidelines: Initial Management of Anaphylaxis using Injection Adrenaline by ANMs (2018). | 5 |
| IV Paracetamol (Acetaminophen) by Age and Weight | US FDA prescribing information, acetaminophen injection (DailyMed, May 2024), section 2: dosage and administration. | 5 |
| Enoxaparin Dose by Indication and Kidney Function | US FDA prescribing information, enoxaparin sodium injection (DailyMed, June 2024), section 2.2 and Table 1 (creatinine clearance below 30 mL/min). | 6 |
| Gentamicin Dose by Age and Weight | US FDA prescribing information, gentamicin sulfate injection (DailyMed, November 2023), dosage and administration: patients with normal renal function. | 5 |
| Ceftriaxone Dose by Indication | US FDA prescribing information, ceftriaxone sodium for injection (DailyMed, January 2026), dosage and administration. | 4 |
| IV Vancomycin Dose by Age | US FDA prescribing information, vancomycin hydrochloride for injection (DailyMed, January 2026), dosage and administration. | 4 |
| IV Acyclovir Dose by Infection and Age | US FDA prescribing information, acyclovir sodium injection (DailyMed, July 2025), dosage and administration. | 4 |
| Oral Amoxicillin by Infection and Weight | US FDA prescribing information, amoxicillin (DailyMed, December 2023), section 2.2, Table 1, and section 2.3. | 4 |
| Oral Azithromycin for Children by Infection | US FDA prescribing information, azithromycin (DailyMed, March 2026), section 2: paediatric patients. | 3 |
| Cefixime Dose for Children and Adults | US FDA prescribing information, cefixime (DailyMed, February 2024), sections 2.1 and 2.2. | 3 |
| Metronidazole for Amoebiasis | US FDA prescribing information, metronidazole tablets (DailyMed, March 2026), dosage and administration: amebiasis. | 2 |
| Albendazole for Neurocysticercosis or Hydatid Disease | US FDA prescribing information, albendazole tablets (DailyMed, September 2019), section 2, Table 1. | 3 |
| Oseltamivir for Influenza by Age and Weight | US FDA prescribing information, oseltamivir phosphate (DailyMed, May 2024), section 2 and Table 1. | 5 |
| Oral Doxycycline for Children and Adults | US FDA prescribing information, doxycycline hyclate (DailyMed, August 2026), dosage and administration. | 3 |
| Co-trimoxazole (Sulfamethoxazole–Trimethoprim) by Weight | US FDA prescribing information, sulfamethoxazole and trimethoprim tablets (DailyMed, December 2024), dosage and administration. | 3 |
| IV Ondansetron by Weight | US FDA prescribing information, ondansetron injection (DailyMed, January 2024), dosage and administration. | 4 |
| Infusion Rate from a Weight-Based Dose | Unit conversion: rate (mL/h) = dose (µg/kg/min) × weight (kg) × 60 ÷ concentration (µg/mL). It implies no dose for any drug. | 2 |
| IV Drip Rate in Drops per Minute | Unit conversion: drops per minute = volume (mL) × drop factor (drops/mL) ÷ time (minutes). It implies no volume or rate for any fluid. | 2 |
| Morphine Milligram Equivalents per Day (CDC 2022) | Dowell D et al. CDC Clinical Practice Guideline for Prescribing Opioids for Pain, United States, 2022. MMWR Recomm Rep 2022;71(3):1-95 (Table: conversion factors adapted from Von Korff 2008 and Nielsen 2016). | 3 |
How accuracy is measured
The accuracy figures on the home page and the evaluation page come from a completed run of the method below, published with its raw outputs so anyone can re-score it. The provisional figures in the changelog entry of 31 October are from internal runs, partly on question sets outside this method (MedXpertQA and HealthBench), and their raw outputs are not published. The results are on the evaluation page.
Public question sets
MedQA (USMLE-style, four options), the validation split of MedMCQA (AIIMS and NEET-PG questions) and PubMedQA, used as published, with no selection.
Clinical cases with a marking scheme
Free-text questions on label dosing, contraindications, interactions and calculations, each graded by code against the label sentence it comes from. Every PubMed identifier and DOI an answer cites is looked up to see whether it exists.
Official sources
Yes-or-no questions on CDSCO prohibitions, the National List of Essential Medicines and recent FDA safety communications, built from the documents themselves, with a matching control for each kind so that always answering yes cannot score.
One harness for every system
Ayulex runs its production pipeline; comparison models receive the same question text through the same script on the same date, with the cost per question recorded.
The evaluation page has the method in full and the calculator results.
Known limitations
Where Ayulex cannot and must not replace clinical judgement.
- Ayulex supports the reasoning of a qualified professional. It does not examine the patient and can be wrong. Every decision must be confirmed by the clinician responsible.
- A statement in an answer without a reference number was not checked against a retrieved source.
- The check shown under an answer is automatic. A separate step compares up to 12 cited sentences in an answer, those with numbers first, with the retrieved excerpt; any further cited sentences are not checked. Doses written as an amount or an amount per kg are matched by number and unit with the material read for the answer, your own messages included; how often a dose is given is not compared, and a dose written as a rate with a slash, such as mg/day, is not read. It can pass a wrong statement or flag a correct one, and it sees only the excerpt that was retrieved, not the whole paper or label.
- Drug labels are US FDA labels. Dosing, formulations and availability can differ in other countries, and national programme regimens take precedence where they apply.
- The CDSCO list reflects each document's date. A court order or notification issued since may change a drug's status; check CDSCO before relying on it in a legal or regulatory matter.
- Jan Aushadhi availability in India is not covered. Indian brand names and MRPs come from NPPA's lists (Pharma Sahi Daam and the anti-cancer price list) as of their stated dates; prices and availability may have changed since.
- ICD-11 codes in an answer are not looked up in the WHO classification. Confirm any code in the WHO browser before recording it.
- The internal protocol and drug summaries have not yet been reviewed by an independent clinician.
- The patient details above an answer are read from the conversation in code, and when the planning model runs it may add a one-line summary of the case. Either can misread the case. Check them first; the answer is built on them.
- The can't-miss checklists are compiled by Ayulex from standard emergency medicine teaching and are pending independent clinical review. They prompt the answer to consider dangerous causes; they are not a differential diagnosis.
- Doses are multiplied out only when the answer gives a mg/kg dose and the weight is in the conversation. The label's maximum is read from its dosing sentences, which can be missed or worded in a way the code does not read; the sentence used is always quoted.
- The support line under an answer summarises the automatic checks. It is not a measured probability that the answer is right, and a well-supported answer can still be wrong.
- A sentence revised after the safety check is rewritten by a model and checked in code for added doses, drugs and citations, not for clinical correctness. Review it like the rest of the answer.
- Interpretation of attached images and reports (ECGs, radiographs, lab reports) is assistive and does not replace formal reporting. DICOM studies are not supported; text extracted from a PDF can misalign table columns.
- Photos and scanned pages are sent as images, with any names or numbers they show. Identifiers with a recognisable format are removed only from typed text and from text read out of a PDF, so crop or cover them before attaching an image.
- Laboratory reference ranges vary between laboratories; the reporting laboratory's range takes precedence over the table used here.
- Paediatric and neonatal dosing needs specialist confirmation.
- In a time-critical emergency, follow your resuscitation algorithm; do not wait for software.
- The role you select is self-declared. Professional registration is not yet verified.
Report an error
If you find a wrong dose, a guideline conflict, a calculation you disagree with or a citation that does not support its sentence, tell us through the contact form. Include the question you asked and what was wrong; leave out anything that identifies a patient. Every report is read by the team, and corrections are recorded in the changelog below.
Changelog
Changes to formulas, data sources and the way answers are checked.
November 2026, sixth update · 15 November 2026
- Correction: the dose check reads doses written as an amount or an amount per kg and compares them, by number and unit, with the material read for the answer, your own messages included. It does not compare how often a dose is given, and it does not read a rate written with a slash, such as mg/day or units/h (other than per kg), or a unit spelt out, such as micrograms. This page and the help pages said every dose was looked for.
November 2026, fifth update · 14 November 2026
- Corrections to this page. The internal protocol and drug summaries have not yet been reviewed by an independent clinician; this page said that review was in progress.
- The weekly benchmark added on 18 October has run only when started by hand since 29 October, so that nothing is charged unasked. No reviewed baseline has been set, so the regression check described there does not run.
- Photos and scanned pages are sent as images, with any names or numbers they show; identifiers with a recognisable format are removed only from typed text and from text read out of a PDF. The privacy policy and the limitations above now say so.
- Other pages were corrected in the same way, among them what the language model provider keeps, how passwords are stored (only as a one-way hash in Ayulex's database), how many cited statements the citation check covers (up to 12 in an answer) and when a score was computed in code (when a calculator is shown with it).
- Current setup: the citation check is a separate step by a model from the same provider. For questions that need reasoning, that model is smaller than the one that wrote the answer; for simple questions it is the same model. Searching the evidence library by meaning needs a model from another company, so it is off and the library is searched by words only, although the 18 October entry describes searching by meaning. An answer a model refuses, or that a content filter stops, is written or finished by another model from the same provider, not from another company as the 29 October entry says. Since 31 October, PubMed is searched only for a question that asks for guidelines, trials, studies or recent evidence.
- Expert reasoning: weak or partly supported answers no longer offer to ask again with it, because in this setup it has not been shown to give better-supported answers. It can still be turned on from the answer options.
November 2026, fourth update · 12 November 2026
- Calculators: 48 added between 6 and 8 November, each tested against worked examples. Thirty, in a new Drug doses group, work out doses, volumes and rates in code, most from India's programme guidance, published guidelines or US prescribing information. Eighteen are scores: RCRI, sPESI, PERC, NEWS2, ABCD2, bedside Schwartz eGFR, FIB-4, Maddrey, BISAP, Padua, 4Ts, Caprini, SOFA, APACHE II, the NIH Stroke Scale, PHQ-9, GAD-7, and BMI with body surface area and ideal body weight.
- Guideline library: further documents from CDC, NCCIH, UKHSA, NCI, ICMR and NTEP, more of the MHRA's Drug Safety Updates, and openly licensed guidelines and reviews from journals.
- A question sent with a photo or a scanned page is now answered by the model that answers questions needing reasoning; until then the small model answered it in this setup.
November 2026, third update · 5 November 2026
- Indian brand names: medicines sold in India as NPPA lists them on Pharma Sahi Daam are now held, with each product's composition, pack, company and MRP. A brand named in a question (such as Glycomet GP or Jalra) is read as the medicines it contains, so the label, interaction and banned-list checks apply to it, and its products are given with NPPA named as the source. A short brand name is read only with its strength written after it (Dolo 650, Pan 40). The list can be searched by medicine or brand at Medicine prices, India.
November 2026, second update · 4 November 2026
- Generic names used in India: 174 names on India's own lists that were not recognised before are now read as the drug they are, among them thiopentone, dextropropoxyphene, gliclazide, vildagliptin, bilastine and 5-fluorouracil. They come from NLEM 2022, CDSCO's prohibited drugs, NPPA's anti-cancer price list and NPPA's retail prices for new drugs in 2014 and since 2019 (the years NPPA's site lists). A drug on CDSCO's prohibited list is now found under either of its names, so a ban on terfenadine or dextropropoxyphene, or on a combination the list spells "Cetrizine", is no longer missed.
- Indian brand names of anti-cancer drugs: the National Pharmaceutical Pricing Authority's list of revised MRPs of non-scheduled anti-cancer drugs, effective 9 November 2026, is now held, row by row. A brand named in a question (such as Zybiraa or Abiratas) is read as the drug it contains, its company, strength, pack and MRP are given with NPPA named as the source, and the whole list can be searched at Cancer drug prices, India. Brands of other medicines are not yet included.
November 2026, first update · 2 November 2026
- India's prohibited and restricted drugs: notifications that appear only on CDSCO's gazette page were read one by one and added, including nimesulide above 100 mg (S.O. 6091(E), 2025), chlorpheniramine with phenylephrine in children below four (S.O. 1717(E), 2025, and S.O. 4595(E), 2026), buclizine as an appetite stimulant, import bans under Section 10A, and bans for animal use. Drafts are not listed as bans.
- FDA Drug Safety Communications: the list now goes back to January 2010, and what each one says reaches the answer, not only its title.
- National List of Essential Medicines: the list's own record of what the 2022 revision added and deleted is now held, so a medicine it removed, such as ranitidine, is described as deleted and not only as absent.
- Calculators: a score of tick boxes shows no result until it has been answered, and sex is asked rather than assumed. Blood urea in mg/dL, as Indian laboratories report it, is now told apart from BUN in the Glasgow-Blatchford score, where it had to be converted by hand. A number that looks like the wrong unit is flagged. The Glasgow Coma Scale takes an intubated patient, and the Parkland formula gives the rate to run now.
- If the connection drops while a signed-in user's answer is being written, the answer is finished on the server and appears when ready, without asking the question again.
- Follow-up questions are now written for the case by the answering model, and shown as choices and fields under the answer.
- Cookies: every visitor is now asked. Google's sign-in script is not requested until they agree.
- Any country can now be chosen. Where Ayulex holds no list of that country's own, Settings says so, and answers name the country's guidance only where it is known. Answers can be written in 16 languages; the interface is in English, and accuracy has been measured in English only.
- Country of practice: answers and checks now follow it. India's banned-drug list and essential medicines apply in India, the MHRA's Drug Safety Update comes first in the UK, and the US withdrawn-drug list and FDA notices apply in the US. Units and exam follow the country unless you set them yourself, and a question that names a country is answered for that country.
October 2026, eleventh update · 31 October 2026
- Reasoning: questions that need reasoning are now written by a stronger model, from a shorter brief. On 200 hard exam questions (MedXpertQA) this scored 56% against 35% for the previous setup; on 60 HealthBench questions, 0.50 against 0.34. These are provisional results on benchmark sets, not a clinical evaluation.
- Fewer model calls: questions are now planned in code, so most answers take one or two calls instead of three or four, and the first words arrive sooner.
- Dose checks: the label of each medicine an answer recommends is now read after it is written, so more doses can be confirmed against a label without a model. A dose found in no source is still marked and listed.
- Exam questions with options are now answered by a larger model with longer reasoning: 71% against 54% on a new sample of 100 hard questions (MedXpertQA), and 62 of 62 official-source questions against 60. Provisional results on benchmark sets, not a clinical evaluation.
- After an answer is written, a dose not found in what was read is looked for in every product form's label for that medicine and in the guideline library, and counts only for its own medicine. A citation whose source does not state the claim is removed, and the change is shown.
- Not run for these answers: the answer contract and the reasoning note. A patient's case line now comes from the details read in code.
October 2026, tenth update · 30 October 2026
- Reference questions without a language model: a first question that only asks what LactMed says about a medicine in breastfeeding, LiverTox's assessment of its liver injury, its entries in CDSCO's prohibited list, its NLEM listing, or an Indian anti-cancer brand's MRP on NPPA's list is answered from that source's own text, and says so. Anything about a patient, a dose or a reason goes to the model, and so does a medicine the data does not list.
- First published evaluation run, with raw outputs, on the evaluation page.
- Models: answers now run on models from one provider. Second opinions, second image reads and the exam panel vote need models from several companies, so they are off in this setup, including where earlier entries below describe them. Sources, calculators, dose checks, label safety checks and the citation check run as before; the citation check is now a separate step by the same model that wrote the answer, which is less independent than a model from another company.
October 2026, ninth update · 29 October 2026
- Refusals: an answer a model declines to write, or that a provider's content filter stops, is not shown. A model from another company writes or finishes it from the same evidence.
- Hard cases: only a question that describes a specific patient can be treated as a hard case. A general teaching question, however detailed, is answered by the standard model.
- Follow-up questions before medicines: asked only when a specific person is described (a patient, or someone asking for themselves or a relative). A general question such as the first-line treatment of a condition is answered at once, covering the cases where the choice differs.
- Stalled answers: if nothing arrives for 35 seconds, the page stops waiting. A signed-in user's answer, which is finished and saved on the server, is loaded; otherwise the question can be asked again.
- Limits renew in 24-hour windows: the first question opens a window, and the questions and Expert reasoning answers counted in it all renew together exactly 24 hours after it opened. Reaching a limit shows when it renews, in the device's own time.
- Cross-checks: independent second opinions and second image reads are limited to a share of model use in each account's 24-hour window. Once it is used, answers say they were written without them, and still carry sources, dose checks, label safety checks and the citation check. Questions and Expert reasoning answers are not reduced.
October 2026, eighth update · 23 October 2026
- Rare diseases: for a case recognised as hard, the findings are matched in code against Orphanet's clinical-feature annotations of 4,355 rare diseases, including findings read from laboratory values. The closest matches go to the answer as cited evidence, to be kept or excluded, and the doctor sees them with the typical features the case does not mention and links to Orphanet.
- Reliability and cost: a second route to the models when the usual one fails, a daily spending cap, time limits that stop abandoned requests, and paid plans that survive a billing outage.
- Breadth: questions about trials or new treatments now retrieve registered trials from ClinicalTrials.gov; complex and in-depth questions retrieve more labels and literature; Ayulex's own evidence library is used as soon as it has content.
- Speed: second opinions start while the question is planned, and are waited for less once one has arrived; optional database reads no longer hold an answer; the citation check runs while missing parts are written. Each stage's time is recorded.
- Reliability: an answer whose model stops part-way, or reaches its length limit, is finished by another model from where it stopped instead of being shown cut short.
- Interface: dialogs keep keyboard focus until closed, controls on phones are easier to tap, and the conversation no longer redraws while an answer arrives.
October 2026, seventh update · 19 October 2026
- Second opinions: for a diagnostic question about a patient, two models from companies other than the one writing the answer each read the case separately and give a differential. The answer must keep or explicitly exclude every diagnosis they raise; the doctor is shown where the models agreed, where they disagreed, and which of their diagnoses the answer discusses.
- Hard cases: when the independent second opinions disagree on the most likely diagnosis, rate nothing as likely, or barely overlap, the case is answered by a stronger model with deeper reasoning and a diagnostic-puzzle method (pivot findings, mechanisms, one diagnosis or two, the confirming test, the next step if tests are unrevealing). PubMed is searched for case reports and reviews of the distinctive findings and of the leading rare diagnosis, and the answer says it was treated as a hard case.
- Images read twice: an attached ECG, radiograph, photo or report is read separately by two vision models from different companies before the answer. Findings both read are used; a value the two read differently is not stated as fact, the answer asks for it to be confirmed, and the doctor is shown which values to check. An image the readers mostly disagree on is answered as a hard case.
- Reused answers: a general reference question that is not about a patient, answered as the first message of a conversation and passing its checks, is answered again from the same checked answer for 6 hours when asked identically with the same settings, at no model cost, and marked as reused. Patient cases are never reused.
- Fewer tokens without less information: label sections the model was already shown in full are not sent to it again, and the calculator list sent with every step is shorter, with full detail for the calculators the question needs.
- Answer structure: every answer about a patient is checked in code for its workup, when to escalate, and a dose table for the medicines it recommends. A missing part is written from the answer itself, may not add a medicine, and is marked as added.
- Answers no longer go missing: an answer whose writing used up its steps on lookups is finished from the evidence already found; a signed-in answer interrupted by a dropped connection is finished and saved; a question left without an answer offers to get one.
October 2026, sixth update · 18 October 2026
- Drug names: thousands more generic names, and US brand names mapped to their ingredients, from NLM's RxNorm. A patient's allergy or medicine written by brand now reaches the safety checks, and an allergy to a named drug now covers its class.
- Quality tracking: each answer records what its checks found, in counts and categories (never its text), linked to any rating it gets, so the weakest areas can be found and fixed.
- Removed: the exploratory answer style (every answer now uses established criteria at low randomness), the cardiology and neurology presets (specialty focus is chosen automatically), and the chart tool. The medication review and SOAP note presets remain.
October 2026, fifth update · 18 October 2026
- Doses for the patient: a mg/kg dose in an answer is now multiplied out in code for the patient's weight, with the daily total, and compared with the label's maximum, which is quoted. An amount the answer states that does not match the multiplication, a dose above the maximum, or a dose 1,000 times off the label is a serious finding, and the sentence is revised.
- Can't-miss check: for a patient presenting with one of thirteen complaints, the answering model is given Ayulex's checklist of time-critical causes and asked to rule each in or out; afterwards code checks the answer for each one by name and lists any it does not address.
- An automated weekly benchmark has been added to compare Ayulex with other models on the same case sets, with a regression check against a reviewed baseline. Results will be published here once runs are reviewed.
October 2026, fourth update · 18 October 2026
- Patient cases now start with a one-line problem representation (age, sex, setting, time course, key findings), written by the planning model in the same call as before, shown above the answer and given to the answering model. No extra model call.
- When a medicine or dose is asked for a patient and details that decide it are missing (age, pregnancy, kidney function, allergies, other medicines), the answer is told by code not to assume them, to say which are missing, and to answer for each case or say there is not enough information.
- Serious safety findings the answer does not address (an allergy to the recommended drug, a contraindication at the patient's kidney function or in pregnancy, a contraindicated interaction, a combination prohibited in India, a withdrawn drug) now lead to the sentences behind them being rewritten from the finding and the label sentence. The change is shown with its reason. A rewrite that adds a dose, a drug or a citation is rejected in code.
- Each checked answer ends with one line on how well it is supported, combined in code from the checks, with the reasons. When it is weak or partly supported and Expert reasoning is available, the answer offers to ask again with it.
- A short first question about one drug ("dose of amoxicillin", "metformin contraindications") is planned in code, saving the planning call. The label is still retrieved and every check still runs.
- Doctors can suggest a correction with a source when they mark an answer wrong. A reviewer checks it; once approved it is given to the answering model, as a cited source, for later questions about the same medicines.
October 2026, third update · 18 October 2026
- FDA Drug Safety Communications since January 2020 are shown under any answer that recommends a drug or class they concern, and the FDA's feeds are read once a day for new ones.
- India: unless another practice country is chosen, each recommended medicine is checked against CDSCO's prohibited and restricted list, matching the exact combination, strength and form, with the notification number, date and current legal status. With India chosen, medicines on the National List of Essential Medicines 2022 are shown with their level of care.
- Breastfeeding and liver: LactMed and LiverTox summaries are retrieved when the case involves breastfeeding or liver disease, and a LiverTox likelihood category A or B is shown as a finding.
- The evidence library is searched by meaning as well as by words, so a question finds passages that use different terms.
- Two case sets added to the evaluation harness: official-sources questions and label-and-calculation cases, both graded by code.
October 2026, second update · 13 October 2026
- Drug labels and literature are now retrieved before the answer is written, from a plan made for each question, so the answer is composed with the evidence in front of the model.
- After an answer is written, each cited sentence is compared with the text of its source by a second model, and each dose is looked up in the retrieved material. The outcome is shown under the answer, including any statement that does not fully match.
- Exam questions with answer options are answered independently by models from different providers, each seeing the options in a different order. A third model is asked only when the first two disagree, and a stronger model only when all three do. The answer is shown with how many agreed; a split decision is marked.
- Calculator requests from the model must state every criterion explicitly; a criterion that is left out is no longer treated as absent.
- Questions are routed by difficulty: patient cases and complex questions get more reasoning effort, straightforward factual questions a faster setting.
- A calculation whose inputs are all given, a single laboratory value and a greeting are answered in code, without a language model; if no model responds, the evidence already retrieved is shown instead of an error.
- Every drug an answer recommends is checked in code against the patient details in the conversation and the drug's own label (kidney function, pregnancy, interactions, allergies), with the label sentence quoted.
- An exam question answered with clear agreement is reused for identical questions for 30 days, and the answer line is always the one the models voted for.
- PDF reports can be attached: text is read in the browser (first 12 pages), scanned PDFs are read as images (first 3 pages), and identifiers are removed from the text before it is sent.
- Saved conversations are kept for 30 days after their last message, then deleted automatically.
- Practice context: choose where you practise (India, United States, United Kingdom or international), laboratory units and target exam. Answers follow that country's guidance where it applies and note where it differs from the US label.
October 2026 · 13 October 2026
- Calculators rebuilt on a single shared definition: 25 calculators, each usable directly and each tested against worked examples. Alvarado and corrected sodium added; SI unit entry added.
- CHA₂DS₂-VASc guidance now applies thresholds without the sex point.
- Drug doses and safety statements are now read from the current FDA label, retrieved live, with a link to the label.
- Literature citations are now built from PubMed records retrieved for the question. References written from the model's memory are no longer shown as links.
- Laboratory value checks now match test names exactly, convert units, and decline to classify when the test, unit or sex-specific range is unknown.
- Benchmark figures and comparisons with other products removed from the site pending a published evaluation.
Clinical decisions must always be confirmed by the attending doctor of record.