Social Welfare Analytics

Social welfare analytics

Projects

Each project follows the same staged framework. The details are working hypotheses, to be confirmed against the programme documents.

1. 01

   Chronic disease

   Mapping the Hypertension Care Cascade: A Systems Dynamics Approach to Implementation Research

   Using AI and survey evidence to find gaps in screening, treatment and control, and to test five-year interventions.

   Project details

   Interventions

   individual

   Reminders, home BP monitoring.

   family

   Family-based adherence support, salt reduction at home.

   society

   Community screening, ASHA follow-up, supply-chain fixes.

   Key feedback loops

   • R Stock-outs raise dropout, fewer patients are controlled, and visible success weakens.
   • R Control success builds community trust, which increases screening.
   • R Workload shortens counselling, which lowers adherence.
   • B A rising uncontrolled backlog triggers supervisor attention.

   R: reinforcing loop. B: balancing loop.

   Documents

   IHCI protocol and reports, NP-NCD operational guidelines, HEARTS technical package, standard treatment protocol, Ayushman Arogya Mandir / health and wellness centre guidance.

   What the model tracks

   Undiagnosed; diagnosed but untreated; on treatment, uncontrolled; on treatment, controlled; lost to follow-up; drug stock; trained staff.

   Flows

   Screening, diagnosis, treatment initiation, achieving control, dropout, re-engagement, drug procurement and consumption.

   Influencing factors

   Drug availability, staff workload, digital app use, adherence, health literacy, salt intake, travel distance, follow-up reminders.

   Data sources

   NFHS-5, LASI, WHO STEPS, NCD portal, IHCI dashboards, facility registers.

   AI and machine learning

   XGBoost or LightGBM for risk and dropout with SHAP; Cox or DeepSurv for time to dropout; Chronos or Prophet for facility BP-control trends; Whisper for voice surveys; SetFit for coding open comments.

   Surveys

   Patients: access, adherence, costs, counselling, trust. Frontline: workload, training, supplies, digital tools. Scales: Implementation Climate, AIM, IAM, FIM.

   Implementation strategies (ERIC)

   Audit and feedback; develop supply and distribution systems; train and educate stakeholders; task-shifting; reminder systems; monitoring.

   Outcomes (RE-AIM)

   Reach: screened and registered. Effectiveness: controlled BP. Adoption: facilities delivering. Implementation: protocol fidelity. Maintenance: year-5 control rate.

2. 02

   Chronic disease

   Diabetes Care in Practice: Modelling Gaps from Screening to Complications

   A systems model of how programme design meets ground reality, with scenarios for individuals, families and communities.

   Project details

   Interventions

   individual

   Self-management education.

   family

   Shared meals, family support.

   society

   Food environment, workplace and school programmes.

   Key feedback loops

   • R Poor control leads to complications, which cause fear and disengagement, which worsen control.
   • B Lifestyle support leads to weight loss, which improves control and slows progression.

   R: reinforcing loop. B: balancing loop.

   Documents

   NP-NCD guidelines, HWC and standard treatment protocols, state diabetes programme documents.

   What the model tracks

   Normoglycaemia; prediabetes; undiagnosed diabetes; diagnosed; controlled; complications (retinopathy, nephropathy, foot); mortality; medicine and test-strip stocks.

   Flows

   Incidence, screening, diagnosis, treatment initiation, glycaemic control, complication onset (with delay), dropout.

   Influencing factors

   Lifestyle support, diet and activity, test availability, cost burden, foot-care knowledge, diabetes distress, provider workload.

   Data sources

   ICMR-INDIAB, NFHS, NCD portal, facility records.

   AI and machine learning

   Gradient boosting with calibration; sequence models on repeated HbA1c or glucose if data exist; optional retinal image models (fine-tuned ViT or ConvNeXt) if the scheme includes eye screening.

   Surveys

   Adherence, diet and activity, foot care, cost, distress; frontline readiness and supplies.

   Link to hypertension

   Shared staff, platforms and supplies: common modules let both projects test shared-resource loops.

3. 03

   Nutrition

   From Policy to Plate: A Systems Model of Malnutrition Programmes and Anganwadi Services

   Linking scheme documents, caregiver and worker feedback, and simulated pathways to better nutrition.

   Project details

   Interventions

   individual

   Counselling.

   family

   Kitchen gardens, ration and cash support.

   society

   WASH, women's groups, convergence with health services.

   Key feedback loops

   • R Infection and malnutrition reinforce each other.
   • R Worker overload leads to missed growth monitoring and late detection, which increases workload.

   R: reinforcing loop. B: balancing loop.

   Documents

   Poshan Abhiyaan and Poshan 2.0 guidelines, Anganwadi service norms, state nutrition mission documents.

   What the model tracks

   Children by age cohort and status: normal; moderate acute malnutrition; severe acute malnutrition; stunted; anaemic (women, adolescent girls); take-home ration stock; trained workers.

   Flows

   Onset, screening, referral, recovery, relapse, ageing between cohorts, ration delivery.

   Influencing factors

   Worker workload, ration supply, feeding practices, household food security, WASH, maternal education, app usability.

   Data sources

   POSHAN Tracker, NFHS, CNNS.

   AI and machine learning

   WHO growth standards for z-scores; tabular models for wasting risk; anomaly detection (Isolation Forest) on tracker records; optional image or video support for length and MUAC (research grade, validated carefully).

   Surveys

   Caregivers: infant and young child feeding indicators, dietary diversity, food insecurity, ration receipt. Anganwadi workers: workload, app usability, training.

4. 04

   Nutrition

   Midday Meal Delivery Under the Lens: Feedback Loops in School Nutrition Implementation

   Understanding how funds, food, attendance and community monitoring interact over five years.

   Project details

   Interventions

   individual

   Child-preference menu rotation.

   family

   Parental participation in monitoring.

   society

   Community monitoring, local procurement, timely fund flow.

   Key feedback loops

   • R Better meals raise attendance, which strengthens community support for quality.
   • R Fund delay lowers quality, which reduces participation and weakens the case for funds.

   R: reinforcing loop. B: balancing loop.

   Documents

   PM POSHAN guidelines, Programme Approval Board minutes, state notifications, social audit reports.

   What the model tracks

   Enrolled children; regular attenders; children receiving meals; food grain stock; kitchen and cooking capacity; funds; nutritional status.

   Flows

   Enrolment, dropout, grain supply, fund release, meals served, incidents.

   Influencing factors

   Cook honorarium, fund delay, menu quality, food safety, community monitoring, social audit coverage.

   Data sources

   PM POSHAN MIS, UDISE+, Programme Approval Board documents.

   AI and machine learning

   Diet optimisation by linear programming for menu adequacy; plate-waste estimation with food-image segmentation (SAM plus a fine-tuned classifier); LLM analysis of social-audit reports; grain demand forecasting.

   Surveys

   Children (age-appropriate, with assent), parents, teachers and cooks (supply, hygiene, delays).

   Ethics

   Surveys with children need ethics approval, parental consent and child assent.

5. 05

   Palliative care

   Strengthening Palliative Care Delivery: A Systems Modelling Study of Access, Caregiving and Support

   Identifying where the programme reaches people, where it does not, and what would help.

   Project details

   Interventions

   individual

   Symptom control.

   family

   Caregiver training and respite.

   society

   Volunteer networks, awareness, opioid regulatory support.

   Key feedback loops

   • R Volunteers raise community visibility, which raises referrals.
   • R Caregiver burnout leads to withdrawal, which lowers quality of care.

   R: reinforcing loop. B: balancing loop.

   What the model tracks

   Estimated population in need; identified; referred; enrolled; receiving home care; trained nurses and volunteers; opioid stock; caregiver burden.

   Flows

   Identification, referral, enrolment, deaths with or without care, volunteer recruitment and attrition.

   Influencing factors

   Opioid availability and licensing, funds, awareness, stigma, volunteer motivation, travel time.

   Data sources

   Programme records, state palliative data, mortality statistics for estimating need.

   AI and machine learning

   Need estimation from mortality and morbidity data; NLP on case notes with strict privacy handling; Whisper or Indic ASR for caregiver interviews; MuRIL or IndicBERT for theme and sentiment analysis.

   Surveys

   Patients and caregivers: IPOS, ESAS, Zarit Burden Interview. Providers: training, opioid access, workload, moral distress.

   Ethics

   Trauma-sensitive protocols; interviews can pause or stop; support referrals arranged.

6. 06

   Animal health

   Containing Foot-and-Mouth Disease and Lumpy Skin Disease in Cattle: Vaccination, Reporting and Farmer Behaviour in a Systems Model

   Simulating outbreak and vaccination dynamics to guide field-level implementation.

   Project details

   Interventions

   individual

   Farmer education.

   family

   Household livestock management.

   society

   Cooperative-based vaccination, movement control, fair compensation.

   Key feedback loops

   • B An outbreak raises awareness, which raises vaccine uptake and reduces spread.
   • R Fear of restrictions causes under-reporting, which causes larger outbreaks.

   R: reinforcing loop. B: balancing loop.

   Documents

   National Animal Disease Control Programme (FMD), LSD control and vaccination guidance, NADRS reporting protocols, Bharat Pashudhan and tagging documents.

   What the model tracks

   Susceptible; vaccinated (immune, waning); exposed; infectious; recovered; dead; vaccine doses; cold-chain capacity; vaccinators. For LSD, a vector population (flies, mosquitoes, ticks).

   Flows

   Vaccination, waning, infection, recovery, culling or death, animal movement, vector growth.

   Influencing factors

   Cold-chain capacity, vaccinator availability, reporting delay, market and movement, compensation, farmer awareness, rainfall and temperature.

   Data sources

   NADRS, vaccination dashboards, ear-tag data, climate data.

   AI and machine learning

   Outbreak prediction with random forest or XGBoost using climate and movement covariates; spatial analysis (geopandas, Moran's I); skin-lesion image classifier for LSD (fine-tuned DINOv2, ViT or YOLO) as a frontline aid; SEIR parameter estimation with Bayesian calibration; optional agent-based village model.

   Surveys

   Farmers: knowledge, attitudes and practices, vaccination barriers, reporting behaviour. Veterinarians and para-vets: cold chain, logistics, workload. Voice interviews in local languages.

   Implementation strategies (ERIC)

   Develop supply and distribution systems; train and educate stakeholders; build a coalition; financial strategies (compensation); monitoring and feedback.