Skip to contents

Inflammatory age (iAge proxy)

Source: vignettes/articles/inflammatory_age.Rmd
inflammatory_age.Rmd

Scope

Compute a simplified inflammatory age proxy (iAge) as a weighted sum of CRP, IL6, and TNFa. Supports NA policies and high-missingness warnings. Units assumed: CRP mg/L; IL6, TNFa pg/mL (no internal conversions).

Important note

This function is a practical proxy and not the published multi-marker iAge model. Use it for exploratory feature engineering and cohort-level analyses, not as a standalone clinical decision metric.

Background

iAge is a simplified implementation of the inflammatory age proxy described by Sayed et al. (2021, Nature Aging). The original model uses a 50+ cytokine proteomics panel; this function computes a weighted sum of three routinely available markers—CRP, IL-6, and TNF-α—as a practical, single-site approximation.

Elevated iAge scores have been associated with:

  • Biological age acceleration independent of chronological age
  • Increased frailty, functional decline, and all-cause mortality risk in longitudinal cohorts
  • Cardiometabolic and neurodegenerative disease incidence
Marker Biological role Default weight
CRP Acute-phase protein; reflects systemic IL-6 signalling 0.33
IL-6 Pro-inflammatory interleukin; master acute-phase regulator 0.33
TNF-α Pleiotropic cytokine; adipose source in metabolic disease 0.34

Approximate reference ranges (healthy adults, standard lab units): - CRP < 3 mg/L (hs-CRP < 1 mg/L = low cardiovascular risk) - IL-6 < 7 pg/mL - TNF-α 1–6 pg/mL

Load packages and demo data

Six participants spanning low to markedly elevated inflammatory burden, with one missing CRP to demonstrate NA handling.

library(HealthMarkers)
library(tibble)

df <- tibble::tibble(
  CRP  = c(0.8, 1.5, 3.5,  8.0,   NA, 42.0), # mg/L
  IL6  = c(1.2, 2.0, 4.1,  6.8,  3.2, 15.0), # pg/mL
  TNFa = c(0.8, 1.0, 1.8,  2.5,  1.4,  3.2)  # pg/mL
)

Column map (required)

Map each marker to your column names; all three are mandatory.

col_map <- list(CRP = "CRP", IL6 = "IL6", TNFa = "TNFa")

Core calculation

Default behavior (na_action = "omit"/ignore/warn): NAs are ignored in the weighted sum (treated as 0 contribution), preserving row count.

ia_default <- iAge(
  data = df,
  col_map = col_map,
  na_action = "omit",
  verbose = FALSE
)

ia_default
#> # A tibble: 6 × 1
#>     iAge
#>    <dbl>
#> 1  0.932
#> 2  1.50 
#> 3  3.12 
#> 4  5.73 
#> 5  1.53 
#> 6 19.9

Interpreting iAge scores

As a weighted sum of (unstandardised) inflammatory markers, the raw score is influenced by the absolute magnitude of CRP (which dominates when elevated). As a broad guide:

Score Inflammatory profile
0–2 Low burden; typical of healthy younger adults
2–8 Sub-clinical elevation; common in overweight, sedentary, or older adults
> 8 High chronic inflammation; consistent with metabolic syndrome or active disease

For cross-cohort comparison, normalise the score with normalize_vec() (method "z" or "robust") before downstream modelling.

Keep NA vs drop

na_action = "keep" returns NA if any required marker is missing in a row. na_action = "error" stops on missing required inputs.

ia_keep <- iAge(
  data = df,
  col_map = col_map,
  na_action = "keep"
)

ia_keep
#> # A tibble: 6 × 1
#>     iAge
#>    <dbl>
#> 1  0.932
#> 2  1.50 
#> 3  3.12 
#> 4  5.73 
#> 5 NA    
#> 6 19.9

Extreme values

Extreme marker values will produce extreme scores. Pre-filter implausible values before calling.

df_ext <- df
df_ext$CRP[2] <- 500  # extreme on purpose
# Pre-filter or cap before passing to iAge
df_ext$CRP[df_ext$CRP > 300] <- 300

ia_filtered <- iAge(
  data = df_ext,
  col_map = col_map,
  verbose = FALSE
)

ia_filtered
#> # A tibble: 6 × 1
#>      iAge
#>     <dbl>
#> 1   0.932
#> 2 100    
#> 3   3.12 
#> 4   5.73 
#> 5  NA    
#> 6  19.9

Weights

  • Default weights: CRP 0.33, IL6 0.33, TNFa 0.34 (sum to 1). Named weights are reordered to CRP/IL6/TNFa.
  • For cohorts where one marker is more informative, adjust weights (ensure they sum to 1):
iAge(df, col_map = col_map, weights = c(CRP = 0.5, IL6 = 0.25, TNFa = 0.25))
#> # A tibble: 6 × 1
#>    iAge
#>   <dbl>
#> 1  0.9 
#> 2  1.5 
#> 3  3.22
#> 4  6.32
#> 5 NA   
#> 6 25.6

Expectations

  • All three markers must be present and numeric; coercion to numeric will error if NAs are introduced.
  • na_action behaviors:
    • omit/ignore/warn: missing markers contribute 0 to the sum; rows preserved.
    • keep: any missing required marker yields NA for that row.
    • error: abort if any required marker is missing/NA.
  • Output: one-column tibble iAge; rows match input unless you subset beforehand.

Verbose diagnostics

Set verbose = TRUE (and healthmarkers.verbose = "inform") to surface three structured messages on each call: preparing inputs, the column map, and a results summary.

old_opt <- options(healthmarkers.verbose = "inform")
df_v <- tibble::tibble(CRP = 1.2, IL6 = 2.0, TNFa = 1.0)
iAge(df_v, col_map = list(CRP = "CRP", IL6 = "IL6", TNFa = "TNFa"), verbose = TRUE)
#> iAge(): reading input 'df_v' — 1 rows × 3 variables
#> iAge(): col_map (3 columns — 3 specified)
#>   CRP               ->  'CRP'
#>   IL6               ->  'IL6'
#>   TNFa              ->  'TNFa'
#> iAge(): computing markers:
#>   iAge  [weighted sum: CRP*0.33 + IL6*0.33 + TNFa*0.34]
#> iAge(): results: iAge 1/1
#> # A tibble: 1 × 1
#>    iAge
#>   <dbl>
#> 1  1.40
options(old_opt)

Column recognition (multi-biobank)

CRP is widely available in clinical databases (often labelled CRP, hs_CRP, hsCRP, c_reactive_protein). IL-6 and TNF-α require targeted immunoassay panels (e.g., Meso Scale Discovery, Luminex) and are absent from most routine clinical extracts.

# Check which inflammation markers are present in your cohort data
hm_col_report(your_data)

For datasets where only CRP is available, pass a single-marker col_map and assign all weight to CRP:

iAge(
  data    = your_data,
  col_map = list(CRP = "hs_CRP"),  # IL6/TNFa omitted
  weights = c(CRP = 1.0),          # full weight to CRP
  na_action = "keep"
)

See the Multi-Biobank Compatibility article for recognised synonyms across major biobanks.

Tips

  • Keep units consistent (CRP mg/L; IL6/TNFa pg/mL). Convert before calling if needed.
  • Prefer na_action = "keep" when you want to see which rows are incomplete; use omit when you want continuity of the score despite single missing markers.
  • Use verbose = TRUE during setup to see mapping and warnings.
  • Impute upstream (e.g., impute_missing()) if you prefer not to down-weight missing markers to zero.
  • Reference: Sayed N, et al. (2021). An inflammatory aging clock (iAge) based on deep learning tracks multimorbidity, immunosenescence, epigenetic reprogramming, and mortality. Nature Aging, 1, 598–615. https://doi.org/10.1038/s43587-021-00082-y