Research, safety & intended use
Scientific and ethical positioning of O.A.S.I.S.
O.A.S.I.S. is a clinical decision-support prototype for structured screening and referral prioritisation—not a diagnostic system.
01 · Intended use
What O.A.S.I.S. does
A deliberately narrow workflow for clinician-facing screening support.
02 · Boundaries
What O.A.S.I.S. does not do
These limitations are part of the product definition.
Not a diagnostic or autonomous clinical system
- Does not diagnose cancer.
- Does not exclude cancer.
- Does not recommend treatment.
- Does not replace clinical examination.
- Does not autonomously make clinical decisions.
- Is not externally clinically validated.
03 · Hybrid decision model
The highest-urgency signal determines the final recommendation
The hierarchy prevents optional statistical support from downgrading clinical safety rules.
Clinical safety rules
Symptoms, abnormal examinations, and concerning previous findings can raise urgency.
Screening eligibility
Age, pathway applicability, screening history, and configured intervals determine screening status.
Optional statistical support
Experimental statistical support may inform prioritisation but cannot lower rule-derived urgency.
Final recommendation = highest urgency among clinical safety rules, screening eligibility, and optional statistical support
04 · Cancer modules
Oral, breast, and cervical pathways
Each module structures different history and warning-sign inputs while using the same safety-first priority hierarchy.
Oral cancer pathway
Structures oral-screening history, tobacco/areca/alcohol exposure, symptoms, and examination findings. Red-flag findings can raise referral urgency.
Breast cancer pathway
Structures clinical breast-screening history, current warning findings, and high-risk personal or family history to support screening, follow-up, or specialist-risk prioritisation.
Cervical cancer pathway
Structures screening history, previous HPV/VIA/cytology information, HIV status, symptoms, and optional consented research variables. Experimental statistical support cannot lower rule-derived urgency.
05 · Mathematical models
Four compact equations used to explain modelling and evaluation
These equations describe standard statistical concepts; they do not establish clinical validity for O.A.S.I.S.
Logistic regression
Maps a linear combination of input variables to a value between 0 and 1. In O.A.S.I.S., statistical output is experimental support and is not a diagnostic probability.
Sensitivity
The proportion of true positive cases correctly identified by a model or test in an evaluation dataset.
Specificity
The proportion of true negative cases correctly identified by a model or test in an evaluation dataset.
Brier score
Measures squared error between predicted probabilities and observed binary outcomes; lower is better when evaluating calibrated probabilistic predictions.
06 · Research basis
Evidence informs the design; it does not validate this prototype
Each reference separates what the source supports from what O.A.S.I.S. actually implements.
Clinical Decision Support Systems in Breast Cancer: A Systematic Review
Shows that breast-cancer clinical decision-support systems span multiple clinical functions and can assist healthcare professionals with decision-making.
O.A.S.I.S. implements an explainable screening/referral support workflow. This review does not clinically validate O.A.S.I.S.
A Model for Predicting Cervical Cancer Using Machine Learning Algorithms
Demonstrates comparison of conventional machine-learning classifiers on a cervical-cancer risk-factor dataset.
O.A.S.I.S. treats statistical modelling as optional experimental support; published performance from this paper is not claimed for O.A.S.I.S.
Breast cancer risk prediction using machine learning: a systematic review
Reviews machine-learning and deep-learning research using mammography, radiomics, genomics, reports, and clinical information for breast-cancer risk prediction.
The current breast module is history- and rule-based. It does not implement or claim the performance of the reviewed imaging models.
A Deep Learning Mammography-based Model for Improved Breast Cancer Risk Prediction
Provides an example of research combining mammograms and traditional risk factors for breast-cancer risk prediction.
O.A.S.I.S. does not reproduce this mammography model; multimodal imaging remains future work.
WHO guideline for screening and treatment of cervical pre-cancer lesions for cervical cancer prevention, second edition
Provides cervical-screening and treatment recommendations, including differentiated guidance for women living with HIV.
O.A.S.I.S. uses configurable screening-eligibility logic and must still be aligned with local programmes, tests, resources, and referral pathways.
National Programme for Prevention and Control of Non-Communicable Diseases (2023–2030): Operational Guidelines
Provides the programme context for population-based screening, early diagnosis, referral, and management of common NCDs, including oral, breast, and cervical cancers.
O.A.S.I.S. is designed around programme-oriented screening and referral support without assuming every state or facility uses identical protocols.
07 · Limitations
Known scientific, data, and deployment constraints
These constraints should remain visible during research, evaluation, and any future deployment.
The public cervical dataset used for experimental modelling is limited, incomplete, imbalanced, and not necessarily representative of local screening populations.
O.A.S.I.S. has not undergone prospective or external clinical validation.
Dataset and model behaviour may reflect population bias and may not generalise across regions or demographic groups.
Several inputs depend on self-reported patient history and may be incomplete or inaccurate.
Sensitive reproductive, sexual, and health-history data require informed consent, privacy safeguards, and appropriate access controls.
Screening intervals, eligibility criteria, available tests, and referral routes vary across local protocols.
08 · Future roadmap
What would be needed beyond the current prototype
Future capability must be paired with clinical governance, validation, privacy, and workflow evaluation.
Clinician review
Formal clinical and programme-owner review of rules, wording, workflow, and referral recommendations.
Multilingual interface
Support locally appropriate languages while preserving clinical meaning and accessibility.
Offline-first screening camps
Enable resilient workflows for outreach settings with intermittent connectivity.
Referral completion tracking
Track whether referrals were received and completed instead of displaying unverified status.
Locally validated datasets
Evaluate and calibrate models using representative local datasets with subgroup analysis.
Multimodal imaging support
Explore carefully validated imaging support while keeping clinical oversight explicit.
Health-system interoperability
Integrate with appropriate health-information systems using governed, auditable data exchange.