Skip to contents

Calculate Limit of Detection (LOD)

Source: R/lod-loq.R
measure_lod.Rd

Calculates the limit of detection using one of several accepted methods. The method used is explicitly documented in the output.

Usage

measure_lod(
  data,
  response_col,
  method = c("blank_sd", "calibration", "sn", "precision"),
  conc_col = "nominal_conc",
  sample_type_col = "sample_type",
  calibration = NULL,
  k = 3,
  sn_col = NULL,
  noise = NULL,
  sn_threshold = 3,
  ...
)

Arguments

data

A data frame containing the measurement data.

response_col

Name of the response column.

method

Method for LOD calculation:

  • "blank_sd": 3 * SD of blank samples (requires sample_type == "blank")

  • "calibration": 3.3 * sigma / slope from calibration curve

  • "sn": Signal-to-noise ratio method (requires sn_col or noise estimate)

  • "precision": Based on acceptable precision at low concentrations

conc_col

Name of concentration column (for calibration method).

sample_type_col

Name of sample type column. Default is "sample_type".

calibration

Optional measure_calibration object for calibration method.

k

Multiplier for SD. Default is 3 for LOD.

sn_col

Column containing S/N ratios (for "sn" method).

noise

Noise estimate for S/N calculation (alternative to sn_col).

sn_threshold

S/N threshold for LOD (default 3).

...

Additional arguments passed to method-specific calculations.

Value

A measure_lod object containing:

  • value: The LOD value

  • method: Method used

  • parameters: Method-specific parameters

  • uncertainty: Uncertainty estimate (when available)

Details

Blank SD Method

LOD = mean(blank) + k * SD(blank)

Where k is typically 3. This is a simple but widely accepted approach.

Calibration Method

LOD = k * sigma / slope

Where sigma is the residual standard error of the calibration curve and slope is the calibration slope. k is typically 3.3 for LOD.

Signal-to-Noise Method

LOD is the concentration where S/N = threshold (typically 3:1).

Precision-Based Method

LOD is the lowest concentration where precision (CV) meets a specified criterion.

See also

measure_loq() for limit of quantitation, measure_lod_loq() for calculating both together.

Examples

# Create sample data with blanks
data <- data.frame(
  sample_type = c(rep("blank", 10), rep("standard", 5)),
  response = c(rnorm(10, mean = 0.5, sd = 0.1),
               c(5, 15, 35, 70, 150)),
  nominal_conc = c(rep(0, 10), c(10, 25, 50, 100, 200))
)

# LOD from blank SD
measure_lod(data, "response", method = "blank_sd")
#> <measure_lod>
#>   Value: 0.6889
#>   Method: blank_sd
#>   k: 3
#>   Uncertainty: 0.08057
#>   Parameters:
#>     blank_mean: 0.4341
#>     blank_sd: 0.08493
#>     n_blanks: 10

# LOD from calibration curve
cal <- measure_calibration_fit(
  data[data$sample_type == "standard", ],
  response ~ nominal_conc
)
measure_lod(data, "response", method = "calibration", calibration = cal)
#> <measure_lod>
#>   Value: 7.71
#>   Method: calibration
#>   k: 3
#>   Parameters:
#>     sigma: 1.783
#>     slope: 0.7634