Adipose insulin sensitivity indices

Scope

Fasting adipose insulin sensitivity/resistance indices in one call: Revised_QUICKI, McAuley_index, Adipo_inv, Belfiore_inv_FFA, TyG_inv, TG_HDL_C_inv, and sex-specific VAI/LAP (inverted so more negative = worse). Inputs are fasting; common glucose/insulin/TG/HDL unit conversions are handled internally.

Important: Understanding Inverted Indices

Most indices returned by adipo_is() are originally described in the literature as insulin RESISTANCE markers (higher values = worse metabolic health). To align with the package’s sensitivity-centric philosophy, we have algebraically inverted these indices so that more negative values indicate worse insulin resistance (equivalently, lower insulin sensitivity).

Inversion Strategy

For indices that measure insulin resistance (IR), we apply the transformation:

$\text{Index}_{\text{inv}} = -\text{Index}_{\text{IR}}$

This ensures a consistent interpretation: more negative = worse adipose insulin sensitivity.

Examples of Inversions

TG_HDL_C_inv = −(TG / HDL)
Original resistance index: TG/HDL (higher = worse IR)
TyG_inv = −ln(TG × G0 / 2)
Original resistance index: TyG (higher = worse IR)
LAP_inv (sex-specific) = −(LAP values)
Original: Lipid Accumulation Product (Kahn 2005)
VAI_inv (sex-specific) = −(VAI values)
Original: Visceral Adiposity Index (Amato 2010)
Adipo_inv = −(FFA × I0)
Original: FFA × Fasting Insulin (higher = worse sensitivity)
Belfiore_inv_FFA = −2 / (I0 × FFA + 1)
Original: Belfiore index for insulin sensitivity using FFA

Indices NOT Inverted

Revised_QUICKI and McAuley_index are retained in their original orientation because they are already described as insulin SENSITIVITY indices (higher values = better sensitivity). No inversion is applied.

Practical Interpretation

In plots and models: - More negative values of *_inv indices → worse adipose insulin sensitivity - More positive values of Revised_QUICKI or McAuley_index → better insulin sensitivity

When combining multiple indices for clustering or PCA, use normalize = "z" to put all indices on a comparable scale before analysis.

When to use this

You need multiple adipose IR markers for metabolic phenotyping or clustering.
You have fasting glucose, insulin, lipids, and adiposity (waist/BMI); sex is optional but improves VAI/LAP labeling.
You want built-in plausibility screening and NA policy controls.

What you need

Required columns: G0 (mmol/L), I0 (pmol/L), TG (mmol/L), HDL_c (mmol/L), FFA (mmol/L), waist (cm), bmi (kg/m^2).
Fasting values only; post-prandial inputs will misstate IR.
Units before conversion: glucose mmol/L, insulin pmol/L, TG mmol/L, HDL mmol/L, FFA mmol/L, waist cm, BMI kg/m^2. If different, convert first.
Optional but helpful: sex (1/2 or labels) to understand which VAI/LAP variant applies.

Load packages and data

Use a small subset of the simulated data. Swap sim_small for your own data.

library(HealthMarkers)
library(dplyr)

sim_path <- system.file("extdata", "simulated_hm_data.rds", package = "HealthMarkers")
sim <- readRDS(sim_path)
sim_small <- dplyr::slice_head(sim, n = 30)

Column map

Map required inputs; left-hand keys are fixed.

col_map <- list(
  G0 = "G0",
  I0 = "I0",
  TG = "TG",
  HDL_c = "HDL_c",
  FFA = "FFA",
  waist = "waist",
  bmi = "BMI"
)

Walkthrough (compute and read results)

adipo_is() returns a tibble of only the computed index columns — not the original data. Bind back to your source rows with dplyr::bind_cols() if needed.

ais <- adipo_is(
  data = sim_small,
  col_map = col_map,
  na_action = "keep",
  normalize = "none"
)

# Group 1 — non-sex-specific indices
ais %>%
  slice_head(n = 5) %>%
  select(Revised_QUICKI, McAuley_index, Adipo_inv, Belfiore_inv_FFA, TyG_inv)
#> # A tibble: 5 × 5
#>   Revised_QUICKI McAuley_index Adipo_inv Belfiore_inv_FFA TyG_inv
#>            <dbl>         <dbl>     <dbl>            <dbl>   <dbl>
#> 1          0.338          4.92     -8.06           -0.221   -9.67
#> 2          0.444         10.2      -1.79           -0.716   -8.07
#> 3          0.348          7.18     -7.64           -0.232   -8.43
#> 4          0.369          7.84     -5.59           -0.303   -8.23
#> 5          0.394          5.62     -5.06           -0.330   -8.51

# Group 2 — sex-specific and ratio indices
ais %>%
  slice_head(n = 5) %>%
  select(TG_HDL_C_inv, VAI_Men_inv, VAI_Women_inv, LAP_Men_inv, LAP_Women_inv)
#> # A tibble: 5 × 5
#>   TG_HDL_C_inv VAI_Men_inv VAI_Women_inv LAP_Men_inv LAP_Women_inv
#>          <dbl>       <dbl>         <dbl>       <dbl>         <dbl>
#> 1        -5.55      -3.04          -4.61     -106.          -128. 
#> 2        -1.35      -0.783         -1.19      -25.0          -30.0
#> 3        -1.95      -0.991         -1.51      -19.3          -26.8
#> 4        -1.43      -0.815         -1.24      -24.0          -30.5
#> 5        -2.36      -1.10          -1.67       -9.84         -21.3

Values ending in _inv are inverted so more negative = worse adipose IR. Both sex variants (VAI_Men/Women, LAP_Men/Women) are always returned; pick the appropriate column per subject based on biological sex:

# Bind sex back from source, then derive a single VAI and LAP per subject.
# Adjust the if_else condition to match your sex coding (e.g. 1/2, "M"/"F").
ais_with_sex <- dplyr::bind_cols(dplyr::select(sim_small, sex), ais)

ais_with_sex %>%
  dplyr::mutate(
    VAI_inv = dplyr::if_else(sex == 1, VAI_Men_inv, VAI_Women_inv),
    LAP_inv = dplyr::if_else(sex == 1, LAP_Men_inv, LAP_Women_inv)
  ) %>%
  dplyr::select(sex, Revised_QUICKI, TyG_inv, VAI_inv, LAP_inv) %>%
  slice_head(n = 5)

Missing data handling

Compare NA policies on a small slice.

demo <- sim_small[1:6, c("G0", "I0", "TG", "HDL_c", "FFA", "waist", "BMI")]
demo$G0[2] <- NA        # missing glucose

ais_keep <- adipo_is(
  data = demo,
  col_map = col_map,
  na_action = "keep",
  normalize = "none",
  verbose = FALSE
)

ais_omit <- adipo_is(
  data = demo,
  col_map = col_map,
  na_action = "omit",
  normalize = "none",
  verbose = FALSE
)

list(
  keep_rows = nrow(ais_keep),
  omit_rows = nrow(ais_omit),
  keep_preview = ais_keep %>% select(ends_with("_inv")) %>% slice_head(n = 3)
)
#> $keep_rows
#> [1] 6
#> 
#> $omit_rows
#> [1] 5
#> 
#> $keep_preview
#> # A tibble: 3 × 7
#>   VAI_Men_inv VAI_Women_inv TG_HDL_C_inv TyG_inv LAP_Men_inv LAP_Women_inv
#>         <dbl>         <dbl>        <dbl>   <dbl>       <dbl>         <dbl>
#> 1      -3.04          -4.61        -5.55   -9.67      -106.         -128. 
#> 2      -0.783         -1.19        -1.35   NA          -25.0         -30.0
#> 3      -0.991         -1.51        -1.95   -8.43       -19.3         -26.8
#> # ℹ 1 more variable: Adipo_inv <dbl>

Normalise outputs

Use normalize to put all indices on a common scale before modelling or dimension reduction:

ais_z <- adipo_is(
  data      = sim_small,
  col_map   = col_map,
  na_action = "keep",
  normalize = "z"      # z-score: mean 0, SD 1
)

ais_z %>%
  slice_head(n = 5) %>%
  select(Revised_QUICKI, McAuley_index, TyG_inv, Adipo_inv)
#> # A tibble: 5 × 4
#>   Revised_QUICKI McAuley_index TyG_inv Adipo_inv
#>            <dbl>         <dbl>   <dbl>     <dbl>
#> 1        -0.442         -0.721  -1.37     0.0702
#> 2         1.38           2.34    1.49     1.06  
#> 3        -0.260          0.575   0.835    0.137 
#> 4         0.0915         0.954   1.20     0.459 
#> 5         0.519         -0.318   0.692    0.543

Available options: "none" (default — raw scale), "z" (mean 0 / SD 1), "range" (0–1 min–max), "robust" (median / IQR), "inverse" (1/x).

Verbose diagnostics

Set verbose = TRUE to emit three structured messages per call:

Preparing inputs — start-of-function signal.
Column map — confirms which data column each required key resolved to. Example: adipo_is(): column map: G0 -> 'G0', I0 -> 'I0', TG -> 'TG', ...
Results summary — shows how many rows computed successfully (non-NA) per output column. Example: adipo_is(): results: Revised_QUICKI 28/30, VAI_Men_inv 28/30, McAuley_index 27/30, ...

verbose = TRUE emits at the "inform" level; you also need options(healthmarkers.verbose = "inform") (or "debug") active in the session for the messages to print:

old_opt <- options(healthmarkers.verbose = "inform")

invisible(adipo_is(
  data          = sim_small[1:5, ],
  col_map       = col_map,
  na_action     = "keep",
  normalize     = "none",
  verbose       = TRUE
))
#> adipo_is(): reading input 'data' — 5 rows × 519 variables
#> adipo_is(): col_map (7 columns — 7 specified)
#>   G0                ->  'G0'
#>   I0                ->  'I0'
#>   TG                ->  'TG'
#>   HDL_c             ->  'HDL_c'
#>   FFA               ->  'FFA'
#>   waist             ->  'waist'
#>   bmi               ->  'BMI'
#> adipo_is(): computing markers:
#>   Revised_QUICKI       [G0, I0, FFA]
#>   VAI_Men_inv          [waist, bmi, TG, HDL_c]
#>   VAI_Women_inv        [waist, bmi, TG, HDL_c]
#>   TG_HDL_C_inv         [TG, HDL_c]
#>   TyG_inv              [TG, G0]
#>   LAP_Men_inv          [waist, TG]
#>   LAP_Women_inv        [waist, TG]
#>   McAuley_index        [I0, TG]
#>   Adipo_inv            [FFA, I0]
#>   Belfiore_inv_FFA     [I0, FFA]
#> adipo_is(): results: id 5/5, Revised_QUICKI 5/5, VAI_Men_inv 5/5, VAI_Women_inv 5/5, TG_HDL_C_inv 5/5, TyG_inv 5/5, LAP_Men_inv 5/5, LAP_Women_inv 5/5, McAuley_index 5/5, Adipo_inv 5/5, Belfiore_inv_FFA 5/5

options(old_opt)  # restore

Enable globally for an entire session: options(healthmarkers.verbose = "inform"). Reset with options(healthmarkers.verbose = NULL) or options(healthmarkers.verbose = "none").

Expectations

All seven required mappings must be provided; missing mappings or columns abort.
na_action controls row handling: keep propagates NA outputs; omit drops rows with missing required inputs; error aborts.
Built-in conversions: glucose 18 (mg/dL), insulin /6 (muU/mL), TG 88.57 (mg/dL), HDL *38.67 (mg/dL); FFA/waist/BMI unchanged.

Common pitfalls

Non-fasting samples inflate TG/glucose and distort TyG/VAI/LAP.
Mis-specified units (e.g., mg/dL glucose without converting) will skew every index; convert upstream if your lab units differ.
Missing sex does not block computation but only one of VAI/LAP applies per subject; interpret accordingly.
For modelling, use normalize = "z" (z-score), "range" (0–1), or "robust" (median/IQR); default "none" keeps raw index scales. See the Normalise outputs section above. ## Validation notes
Revised_QUICKI: typical adults ~0.3–0.6; lower is worse IR. Computed from log10(I0 [mU/L]) + log10(G0 [mg/dL]) + log10(FFA [mmol/L]).
McAuley_index: typical adults ~3–10; higher is better insulin sensitivity. Uses ln(TG [mmol/L]) and ln(I0 [mU/L]) — TG is kept in mmol/L before the internal mg/dL conversion.
TyG_inv: typically −8 to −12 for adults; more negative = worse IR. Computed from TG and G0 both in mg/dL.
LAP_Men_inv / LAP_Women_inv: computed as −(WC−threshold) × TG [mmol/L] (Kahn 2005). Typical values −10 to −80; more negative = worse.
VAI_Men_inv / VAI_Women_inv: uses TG and HDL in mmol/L with the Amato 2010 reference constants (1.03, 1.31, 0.81, 1.52). Typical range −1 to −5; more negative = worse.
Spot-check TG_HDL_C_inv: should equal −(TG_mg/dL / HDL_mg/dL).