How Health-GPS models a person

Global Health Policy Simulation model

Last updated: August 2026

About this note

This guide is the person-centric view of Health-GPS. If you open one modelling page and want the overall gist of the framework, start here.

Health-GPS builds a virtual population of Person entities and updates their characteristics each simulated year. The pages below walk through every main attribute on that person: where it lives in memory, which module sets it, which inputs drive it, and what can change later.

I wrote it for modellers and engineers who need a clear map without reading the whole codebase first. Code pointers are included so you can jump into the source when you need detail.

Topic	Document
Module pipeline (short)	Models overview
Module and ModelName I/O	Simulation models reference
FINCH income, PA, predictors	FINCH guide
Height by income stratum	Height CSV quintile plan
Weight by income stratum	Weight quintile plan
Income categories / strata	Dynamic income categories, Income quintile plan
Person IDs across scenarios	Same person ID plan
Config layout	User Guide, Schemas
Architecture	Software Architecture
Technical index	technical/README.md

What a person is
Assignment equations
When attributes are set
Person ID
Age and gender
Region, sector, and ethnicity
SES and income
Physical activity
Height, weight, and BMI
Other risk factors
Diseases
Alive, death, and migration
HLM France vs FINCH
Where to look in code and config

1. What a person is

Every simulated individual is a Person (src/HealthGPS/person.h). Think of it as the full state vector for one life course inside one scenario run.

The Mermaid diagram below is the overview used on the documentation home page as well. Each box names an attribute group and the core assignment equation. Section 2 writes those equations out in full, matching the C++ implementation.

%%{init: {"flowchart": {"nodeSpacing": 36, "rankSpacing": 48, "padding": 16}}}%%
flowchart TB
    subgraph ROW1 [" "]
        direction LR

        subgraph DEMO ["1. Demographics"]
            direction LR
            AGE["Age<br/>n from population shares<br/>yearly: age := age + 1"]
            GEN["Gender<br/>age-sex table / SRB<br/>male = 1, female = 0"]
            REG["Region / ethnicity<br/>CDF from prevalence"]
            SEC["Sector<br/>Bernoulli rural prevalence"]
        end

        subgraph SOCIO ["2. Socio-economic"]
            direction LR
            SES["SES<br/>ses ~ Normal(mu, sigma)<br/>newborns only on update"]
            INC["Continuous income<br/>I = Z + eps, then clamp<br/>equal-rank categories"]
            CAT["Categorical income<br/>softmax logits to category"]
        end
    end

    PERSON(["PERSON<br/>virtual individual state<br/>src/HealthGPS/person.h"])

    subgraph ROW2 [" "]
        direction LR

        subgraph RF ["3. Behaviour and risk factors"]
            direction LR
            PA["Physical activity<br/>simple: mu * exp(eps - 0.5 sigma^2)<br/>or continuous: clamp(Z + eps)"]
            FOOD["Foods / nutrients<br/>Stage 1: logistic P(zero)<br/>Stage 2: mu * BoxCox^-1(Z, lambda)"]
        end

        subgraph BODY ["4. Body, disease, and status"]
            direction LR
            WHB["Weight / Height / BMI<br/>W = W_exp * q(EI/PA)<br/>BMI = W / h_m^2"]
            DIS["Diseases and death<br/>P = rate * RR / mean RR<br/>P_death = 1 - survival"]
        end
    end

    DEMO --> PERSON
    SOCIO --> PERSON
    PERSON --> RF
    PERSON --> BODY

How Health-GPS models a person. Top row: demographics and socio-economic status side by side. Centre: Person. Bottom row: behaviour/risk factors and body/disease/status side by side. Section 2 writes the equations in full.

Group	Fields on `Person`	Typical owner
Identity	`id()`	`Population`
Core demographics	`age`, `gender`, `region`, `ethnicity`	Demographic module
Urban / rural	`sector`	StaticLinear (when configured)
SES noise	`ses`	SES module
Income	`income_continuous`, `income`, `income_adjustment_stratum`	StaticLinear + `project_requirements`
Physical activity	`physical_activity` (+ often `risk_factors["PhysicalActivity"]`)	StaticLinear or hierarchical RF
Risk factors	`risk_factors` map	Static then dynamic RF models
Anthropometrics	`Height`, `Weight`, `BMI` inside `risk_factors`	Kevin Hall (FINCH) or HLM/EBHLM (France)
Diseases	`diseases` map	Disease module
Status	alive / emigrated / event times	Demographic + migration

Many nutrients and behaviours exist only inside risk_factors. Dedicated members (physical_activity, income_continuous) exist when the model needs them outside that map as well.

2. Assignment equations

These are the equations Health-GPS actually evaluates when assigning person characteristics. Notation matches the code (evaluate_linear_model, inverse_box_cox, update_age_and_death_events, Kevin Hall helpers, and the default disease model).

Shared linear predictor

Used for continuous income, continuous PA, Box-Cox risk factors, and logistic Stage 1 scores:

Z = intercept + sum_k (beta_k * x_k) + sum_m (beta_m * log(x_m))

evaluate_linear_model skips metadata coefficients such as stddev, min, max, and lambda. Predictors x_k come from person fields and risk_factors (age, gender encodings, region/ethnicity dummies, income, nutrients, and so on). See the FINCH guide for predictor name mapping.

Person ID

initial cohort:   id(i) = i + 1          for slot index i = 0 .. N-1
new entrants:     id = next_person_id++  (newborns and immigrants)

IDs are never reused after death or emigration.

Age and gender

n_age,sex = round( population_share(year, age, sex) * virtual_population_size )

Yearly for survivors:

age := age + 1

Birth sex uses the life-table sex ratio at birth (SRB). Encoding used elsewhere:

gender_value(male) = 1
gender_value(female) = 0

Region and ethnicity (optional)

With prevalence tables enabled:

U ~ Uniform(0, 1)
region = first category whose cumulative prevalence(age, gender) >= U

U ~ Uniform(0, 1)
ethnicity = first category whose cumulative prevalence(age_group, gender, region) >= U

Age group is Under18 if age < 18, else Over18. Newborns require an exact age_0 region row.

Sector (StaticLinear, optional)

U ~ Uniform(0, 1)
sector = rural  if U < rural_prevalence(age_group, gender)
       = urban  otherwise

At age 18, if currently rural:

p_urban = 1 - rural_prevalence(Over18, gender) / rural_prevalence(Under18, gender)
U ~ Uniform(0, 1);  become urban if U < p_urban

SES equation

ses ~ Normal(mu, sigma)

from modelling.ses_model (function_name = "normal", parameters [mu, sigma]). Redrawn only for newborns on yearly update.

Continuous income (FINCH)

I0 = Z_income
eps  ~ Normal(0, sigma_income)     if stddev coefficient present
I  = clamp(I0 + eps, min, max) if min/max coefficients present

Store income_continuous = I and risk_factors["income"] = I.

Equal-rank reporting categories with C categories (project_requirements.income.categories):

sort active incomes I_(1) <= ... <= I_(n)
threshold_j = I_( round((n-1) * j/C) )     for j = 1 .. C-1
category(I) = strata[j] if I <= threshold_j, else highest stratum

Adjustment strata (0 .. N-1) use the same equal-rank idea with N = adjustment_income_stratum_count.

Categorical income (India-style)

logit_c = intercept_c + Sum beta_{c,k} * x_k
p_c = exp(logit_c) / Sum_j exp(logit_j)
category ~ Categorical(p)

Physical activity

Simple type:

mu = expected PhysicalActivity(gender, age)
eps ~ Normal(0, sigma)
PA = clamp( mu * exp(eps - 1/2 sigma^2) )

Continuous type:

PA = clamp( Z_PA + eps ),   eps ~ Normal(0, sigma)

Both write physical_activity and risk_factors["PhysicalActivity"].

Foods and other StaticLinear risk factors

Correlated residuals first (r from Cholesky of the residual correlation), then optional two-stage path.

Stage 1 (if a logistic model exists for factor f):

p_zero = logistic(Z_logistic)
U ~ Uniform(0, 1)
if U < p_zero:  risk_factors[f] = 0

Stage 2 (non-zero path, or Box-Cox-only factors):

Z = Z_linear + r * sigma
if |lambda| ~= 0:   BoxCox^{-1}(Z, lambda) = exp(Z)
else:         BoxCox^{-1}(Z, lambda) = (lambda Z + 1)^(1/lambda)     (0 if base <= 0)
risk_factors[f] = clamp( mu_f * BoxCox^{-1}(Z, lambda) )

mu_f is the expected factors-mean value for that person (gender, age, and optional income stratum).

Weight, height, BMI (Kevin Hall)

(EI/PA)_expected = EnergyIntake_expected / PhysicalActivity_expected
(EI/PA)_actual   = EnergyIntake / PhysicalActivity
q = quantile( (EI/PA)_actual / (EI/PA)_expected ; weight quantile curve )
Weight = Weight_expected * q

Height (slope and sigma may depend on income adjustment stratum):

eps_H = Height_residual ~ Normal(0, sigma)   drawn once
Height = Height_expected * (Weight^slope / mean(Weight^slope)) * (exp(eps_H) / exp(1/2 sigma^2))

h_m = Height / 100
BMI = Weight / h_m^2

Diseases

Prevalence initialisation and yearly incidence share the same shape:

RR = product over risk-factor RRs * product over comorbid-disease RRs
P  = rate(age, sex) * RR / mean_RR(age, sex)
U ~ Uniform(0, 1);  disease becomes active if U < P

rate is prevalence at init and incidence on yearly update. Intervention runs may apply PIF:

P := P * (1 - PIF(age, sex, years_since_start))

Remission (if modelled):

U ~ Uniform(0, 1);  status := free if U < remission(age, sex)

Death

Residual (non-modelled) mortality is calibrated from life-table death rates and mean excess-mortality products. For an active person each year:

survival = (1 - residual_death(age, sex))
for each active disease d:
    survival = survival * (1 - excess_mortality_d)
P_death = 1 - survival
U ~ Uniform(0, 1);  die if age >= max_age or U < P_death

Survivors then take age := age + 1.

3. When attributes are set

Attributes are not all drawn once. Health-GPS uses a fixed module order at initialisation and again each year.

flowchart TB
    subgraph init [Initialisation once per run]
        P0[Population creates slots and IDs]
        D0[Demographics: age, gender, region, ethnicity]
        S0[SES draw]
        R0[Static RF model then dynamic generate]
        X0[Diseases: prevalence start]
        A0[Analysis snapshot]
        P0 --> D0 --> S0 --> R0 --> X0 --> A0
    end

    subgraph year [Each simulated year]
        D1[Demographics: death, age++, births]
        M1[Net immigration]
        S1[SES for newborns only]
        R1[Static then dynamic RF update]
        X1[Disease remission and incidence]
        A1[Analysis publish]
        D1 --> M1 --> S1 --> R1 --> X1 --> A1
    end

    init --> year

Timing	What typically happens
Population construction	Lifetime-unique `id` for each slot
Demographic init	Age and gender from population tables; optional region and ethnicity
SES init	Continuous `ses` for everyone
Static RF generate	Sector, income, PA, foods/nutrients (model-dependent)
Dynamic RF generate	Height, weight, BMI, energy state (Kevin Hall) or hierarchical updates (EBHLM)
Disease init	Starting prevalence
Yearly demographic	Deaths, survivors age by one year, births at age 0
Yearly RF update	Newborns re-initialised; children/adults follow dynamic rules
Yearly disease	Remission then new incidence
Migration	Emigration flags or immigrant clones with new IDs

For the module-level I/O view, see Simulation models reference. For the short pipeline summary, see Models overview.

4. Person ID


Field	Private `id_`; read with `id()`. Sentinel `Person::unassigned_id` (`0`) until Population assigns an ID.
Initial cohort	Slot index `i` gets ID `i + 1`.
Newborns and immigrants	Monotonic `next_person_id_++`. Slots may be reused; IDs are never reused.
Inputs	None. Same seed and size give the same initial ID sequence across baseline and intervention.
Later changes	Never. Death and emigration do not free the ID for reuse.

flowchart LR
    CTOR[Population constructor] --> INIT["IDs 1..N"]
    BIRTH[add_newborn_babies] --> NEXT[allocate_next_person_id]
    ADD[add immigrant / entity] --> NEXT
    NEXT --> UNIQUE[Lifetime-unique ID]

Detail: Same person ID plan. Optional CSV export: Individual ID tracking plan.

5. Age and gender

Age


Field	`unsigned int age`
Initial set	`DemographicModule::initialise_population` fills age by sex from datastore population shares for the start year, scaled to the virtual population size.
Inputs	Backend population tables; `inputs.settings` (`country_code`, `size_fraction`, `age_range`).
Each year	Survivors: `age += 1`. Newborns enter at age `0`. People at or above `age_range` upper bound die before ageing further.

Gender


Field	`core::Gender gender`
Initial set	Male and female counts from the same age-sex distribution (with rounding correction on the last bins).
Births	Sex ratio at birth from the life table / birth process; newborns created via `add_newborn_babies(..., gender, ...)`.
Encoding	`gender_to_value`: male = 1, female = 0. FINCH linear models may also use `gender2` (see FINCH guide).
Later changes	Fixed for life.

flowchart TD
    POP[Datastore population by year, age, sex] --> SHARE[Age-sex shares]
    SHARE --> SIZE[Scale to virtual population size]
    SIZE --> ASSIGN[Set person.age and person.gender]
    BIRTHS[Yearly births from birth rates] --> BABY[New person age 0 + sampled sex]
    SURV[Survivors after death process] --> INC[age plus 1]

Code: src/HealthGPS/demographic.cpp (initialise_population, update_age_and_death_events).

6. Region, sector, and ethnicity

These are optional demographics extras. France-style HLM packs often leave them unused. FINCH-style packs usually enable region and ethnicity through project_requirements.demographics.

Region


Field	`std::string region` (default `"unknown"`)
When	Init for everyone; yearly only for newborns (`age == 0`).
How	`initialise_region`: CDF sample from age- and gender-specific prevalence. Exact `age_0` required for newborns; other ages may use closest available age key.
Inputs	Region prevalence CSV registered from the static model file (`RegionFile` path). Gate: `project_requirements.demographics.region`.

Ethnicity


Field	`std::string ethnicity`
When	After region, same timing (init + newborns).
How	`initialise_ethnicity`: prevalence by age group (Under18 / Over18), gender, and current region.
Inputs	Ethnicity CSV (`EthnicityFile`). Gate: `project_requirements.demographics.ethnicity`.
Remap	CSV codes `"1"`..`"4"` become `"ethnicity1"`..`"ethnicity4"`; other names kept as-is.

Sector (urban / rural)


Field	`core::Sector sector`
When	StaticLinear init; yearly update mainly for rural people turning 18.
How	`initialise_sector` uses Under18/Over18 rural prevalence by gender. At age 18, rural people may move to urban with probability derived from the prevalence ratio.
Inputs	`RuralPrevalence` in the static model JSON.

flowchart TD
    PR[project_requirements.demographics] --> GATE{region / ethnicity enabled?}
    GATE -->|yes| REGCSV[Region prevalence CSV]
    GATE -->|yes| ETHCSV[Ethnicity prevalence CSV]
    REGCSV --> REG[person.region]
    REG --> ETH[person.ethnicity depends on region]
    RURAL[RuralPrevalence in static JSON] --> SEC[person.sector]
    SEC --> AGE18[Age 18 rural to urban transition]

Code: demographic.cpp for region/ethnicity; static_linear_model.cpp for sector.

7. SES and income

SES noise and income are different. Mixing them up is a common source of confusion when reading France vs FINCH configs.

SES field


Field	`double ses`
How	`SESNoiseModule`: draw `N(mean, sd)` from `modelling.ses_model` (`function_name` must be `"normal"`).
When	Everyone at init; newborns only on yearly update. Adults keep their draw.
Role	Continuous predictor for hierarchical models (especially HLM France).

Income (FINCH continuous path)

Field	Meaning
`income_continuous`	Continuous income value
`risk_factors["income"]`	Usually the same continuous value in FINCH
`income` (`core::Income`)	Reporting category (Low / LowerMid / UpperMid / High, or 3-category layout)
`income_adjustment_stratum`	Rank bucket `0..N-1` for optional quintile factors-mean tables

Typical order inside StaticLinear generate:

Draw continuous income from the income linear model (+ optional noise), clamp to configured min/max.
Optional overall factors-mean adjustment on income.
Optional equal-rank split into adjustment strata (income_adjustment_stratum).
Optional per-stratum factors-mean on other risk factors / PA.
Equal-rank split into final income categories for outputs (project_requirements.income.categories).

Income (India-style categorical path)

initialise_categorical_income draws a category from logits / softmax and stores the category numeric value in risk_factors["income"]. No continuous income pipeline.

flowchart TD
    SESMOD[modelling.ses_model] --> SES[person.ses]
    INCMODE{Income model type}
    INCMODE -->|continuous FINCH| CONT[income_continuous + RF income]
    INCMODE -->|categorical| CAT[person.income enum]
    CONT --> ADJ[Optional factors-mean]
    ADJ --> STRAT[Optional adjustment strata 0..N-1]
    STRAT --> FINAL[Final income categories 3 or 4 or 5]
    CAT --> OUT[Outputs / predictors]
    FINAL --> OUT

Deep dives: FINCH guide, Dynamic income categories plan, Income quintile factor means plan.

8. Physical activity


Fields	`physical_activity` and usually `risk_factors["PhysicalActivity"]` (kept in sync on the StaticLinear path).
Gate	`project_requirements.physical_activity.enabled`
Simple type	Expected mean by age/sex * lognormal-style noise.
Continuous type	Linear model + noise + clamp to configured bounds.
When	Init for the population; yearly re-init for newborns. Adults are not randomly re-drawn each year on this path.
Kevin Hall	Reads PA when forming energy / PA ratios for weight; it does not own the PA assignment.
HLM France	PA is often just another hierarchical risk factor (`PA`) inside HLM/EBHLM, not the dedicated StaticLinear PA models.

flowchart LR
    PR[project_requirements.physical_activity] --> EN{enabled?}
    EN -->|no| SKIP[Leave unset / unused]
    EN -->|yes| TYPE{type}
    TYPE -->|simple| SIMP[Mean by age sex + noise]
    TYPE -->|continuous| CONT[Linear model + noise + clamp]
    SIMP --> STORE[physical_activity and RF PhysicalActivity]
    CONT --> STORE
    STORE --> KH[Kevin Hall may read PA]

9. Height, weight, and BMI

On FINCH / Kevin Hall packs these are explicit anthropometrics. On HLM France, BMI (and related factors) usually come from the hierarchical model instead.

Weight (Kevin Hall)

Build nutrient intakes and energy intake from foods.
Compare actual energy/PA to expected energy/PA.
Map that ratio onto a weight-quantile curve (optionally one curve per income adjustment stratum).
Set risk_factors["Weight"], then optionally mean-adjust the population.

Children under 19 typically re-initialise weight on update. Adults use the Kevin Hall energy-balance run.

Height (Kevin Hall)

After weight:

Resolve slope / std by sex and optional income stratum (resolve_height_params_for_person).
Draw a lifelong height residual once.
update_height combines expected height, weight slope term, and residual.
Growth updates continue through the child ages configured by the model.

BMI

BMI = Weight / (Height in metres)^2, recomputed after anthropometric updates.

flowchart TD
    FOOD[Foods / nutrients from static model] --> EI[Energy intake]
    PA[Physical activity] --> RATIO[EI/PA actual vs expected]
    EI --> RATIO
    RATIO --> WQ[Weight quantile curve by stratum]
    WQ --> W[Weight]
    W --> H[Height from slope, std, residual]
    W --> BMI[BMI]
    H --> BMI

Detail and console tables: Height CSV quintile plan, Weight quintile plan.

10. Other risk factors

Everything else in Person.risk_factors (foods, nutrients, residuals, policy/trend copies, Kevin Hall internal state, and so on) is owned by the registered static and dynamic models.

StaticLinear (FINCH-style foods and nutrients)

Draw correlated residuals (Cholesky).
Evaluate linear predictors.
Optional two-stage path: logistic chance of zero, then Box-Cox continuous value.
Clamp to configured ranges.
Store value and {name}_residual.
Optional factors-mean adjustment, policies, and trends.

HLM + EBHLM (France-style)

Static HLM walks the configured hierarchy level by level.
Dynamic EBHLM applies age-specific deltas and boundary logic, then factors-mean adjustment where configured.

flowchart TB
    CFG[modelling.risk_factor_models] --> ST[Static model generate]
    CFG --> DY[Dynamic model generate / update]
    ST --> MAP[person.risk_factors]
    DY --> MAP
    MAP --> DIS[Disease relative risks]
    MAP --> OUT[Analysis aggregates and optional ID tracking]

Predictor naming, Box-Cox, and policy equations for FINCH: FINCH guide. Per-model I/O: Simulation models reference.

11. Diseases


Field	`diseases` map of `Disease{status, start_time, time_since_onset}`
Status	`free` or `active`
Initial	Prevalence draw using risk-factor relative risks (and mean RR normalisation).
Yearly	Remission (if applicable), then incidence. Age 0 clears disease history before infant logic runs.
Cancer models	Track `time_since_onset` and duration-based excess mortality.
Inputs	Datastore disease rates and RR tables; diseases selected in `running`; optional PIF on intervention incidence.
Feedback	Active disease excess mortality feeds the demographic death calculation.

flowchart TD
    INIT[initialise_disease_status from prevalence x RR] --> YEAR[Each year]
    YEAR --> REM[Remission draws for active cases]
    REM --> INC[Incidence draws for free cases]
    INC --> EX[Excess mortality tables]
    EX --> DEATH[Demographic death hazard]

Code: disease.cpp, default_disease_model.cpp, default_cancer_model.cpp.

12. Alive, death, and migration

Field / accessor	Meaning
`is_alive()`	False after `die(time)`
`has_emigrated()`	True after `emigrate(time)`
`time_of_death()` / `time_of_migration()`	Event times
`is_active()`	Alive and not emigrated

Death. update_age_and_death_events combines residual (non-modelled) mortality with excess mortality from active diseases. People at max age die. Survivors then age by one year.

Emigration. When net migration for an age-sex cell is negative, active people in that cell are marked emigrated.

Immigration. When net migration is positive, the engine clones a similar active person (partial_clone_entity), assigns a new lifetime ID, and adds them. The clone copies age, gender, region, ethnicity, ses, sector, income category, risk factors, and diseases. Continuous-income extras and some PA/stratum bookkeeping may be incomplete until later module updates.

flowchart TD
    LIFE[Active person] --> HAZ{Death hazard this year?}
    HAZ -->|yes| DEAD[die time]
    HAZ -->|no| AGE[age plus 1]
    LIFE --> NET{Net migration for age-sex}
    NET -->|outflow| EMI[emigrate time]
    NET -->|inflow| CLONE[Clone similar person + new ID]

13. HLM France vs FINCH

Same Person type. Different models fill different fields.

Attribute	HLM France typical path	FINCH typical path
ID, age, gender, death, migration	Shared demographic / population logic	Same
Region / ethnicity / sector	Usually unused	Enabled via demographics + StaticLinear
`ses`	Important HLM predictor	Drawn, but income is the main socio-economic driver
Income continuous / category / stratum	Usually unused	StaticLinear + `project_requirements`
Physical activity	Hierarchical RF (`PA`)	Dedicated PA models + member field
Foods / nutrients	HLM then EBHLM	StaticLinear (+ policies / trends)
Height / Weight / BMI physiology	BMI as hierarchical RF	Kevin Hall anthropometrics
Diseases	Shared disease module	Shared disease module

flowchart LR
    subgraph france [HLM France stack]
        HLM[static hlm] --> EBHLM[dynamic ebhlm]
    end

    subgraph finch [FINCH stack]
        SL[static staticlinear] --> KH[dynamic kevinhall]
    end

    france --> PERSON[Person state]
    finch --> PERSON

Example packs: HLM_France, KevinHall_FINCH.

14. Where to look in code and config

Concern	Primary code	Primary config / data
Person fields	`src/HealthGPS/person.h`	n/a
IDs	`population.cpp`	`running` seeds / population size
Age, gender, births, deaths	`demographic.cpp`	`inputs.settings`, datastore population / births / deaths
Region / ethnicity	`demographic.cpp`	`project_requirements.demographics`, region/ethnicity CSVs
SES	`ses_noise_module.cpp`	`modelling.ses_model`
Sector, income, PA, foods	`static_linear_model.cpp`	static model JSON + CSVs, `project_requirements`
Height, weight, BMI	`kevin_hall_model.cpp`	dynamic Kevin Hall JSON + CSVs
Hierarchical RF (France)	`static_hierarchical_linear_model.cpp`, `dynamic_hierarchical_linear_model.cpp`	HLM / EBHLM JSON
Diseases	`disease.cpp`, default disease/cancer models	datastore disease tables, `running` disease list
Migration clone	`simulation.cpp`	population targets vs simulated counts
Factors-mean adjustment	`risk_factor_adjustable_model.cpp`	`modelling.baseline_adjustments`