Skip to contents

Overview

The measure package includes calculations and plots used in analytical method validation work. The examples reference procedures described in: - ICH Q2(R2): Validation of Analytical Procedures - ISO/IEC 17025: General requirements for testing and calibration laboratories - USP <1225>: Validation of Compendial Procedures - ICH M10: Bioanalytical Method Validation (for applicable workflows)

The examples use simulated measurements to demonstrate calibration, precision, accuracy, uncertainty, and quality-control workflows. They do not establish method suitability or regulatory acceptance; those conclusions depend on the study design, data, and criteria for the intended use.

Calibration Curves

Fitting Calibration Curves

measure_calibration_fit() fits weighted or unweighted calibration curves and returns model diagnostics.

# Create calibration data
set.seed(42)
cal_data <- data.frame(
  nominal_conc = c(1, 5, 10, 25, 50, 100, 250, 500),
  response = c(1, 5, 10, 25, 50, 100, 250, 500) * 1.05 +
             rnorm(8, sd = c(0.1, 0.3, 0.5, 1, 2, 4, 10, 20))
)

# Fit with 1/x^2 weighting (common for bioanalytical methods)
cal <- measure_calibration_fit(
  cal_data,
  response ~ nominal_conc,
  weights = "1/x2"
)

print(cal)
#> <measure_calibration>
#>   Model: linear
#>   Weighting: 1/x2
#>   Formula: response ~ nominal_conc
#>   N points: 8
#>   R²: 0.9989

Visualizing the Calibration

autoplot(cal, type = "curve") +
  labs(
    x = "Nominal concentration (simulated scale)",
    y = "Response (simulated scale)"
  )
Scatter chart of nominal concentration versus response for eight simulated standards, with a fitted weighted regression line. The points follow an increasing trend across the concentration range; the line summarizes the fitted model.

Eight simulated standards and the fitted calibration line from a model weighted by 1/x². The points show the supplied observations; the line is the fitted response over the calibration range.

Checking Residuals

autoplot(cal, type = "residuals") +
  labs(x = "Nominal concentration (simulated scale)")
Residual scatter chart for the eight simulated calibration standards. Nominal concentration is on the x-axis and residual is on the y-axis; the dashed horizontal line marks zero. One point near 250 has a much larger positive residual than the others.

Residuals from the weighted calibration model plotted against nominal concentration, with a horizontal zero reference line. The standard near concentration 250 has a larger positive residual than the other points.

Predicting Unknown Concentrations

unknowns <- data.frame(
  sample_id = c("Sample_1", "Sample_2", "Sample_3"),
  response = c(52.3, 125.8, 280.5)
)

predictions <- measure_calibration_predict(
  cal,
  newdata = unknowns,
  interval = "confidence"
)

cbind(unknowns, predictions)
#>   sample_id response .pred_conc .pred_lower .pred_upper
#> 1  Sample_1     52.3    49.3896     49.3580    49.42121
#> 2  Sample_2    125.8   118.9571    118.9279   118.98627
#> 3  Sample_3    280.5   265.3801    265.3408   265.41938

Calibration Verification

Check the fitted calibration against QC samples:

qc_data <- data.frame(
  sample_id = c("QC_Low", "QC_Mid", "QC_High"),
  nominal_conc = c(3, 75, 400),
  response = c(3.1, 77.5, 395.2)
)

verification <- measure_calibration_verify(cal, qc_data)
print(verification)
#> 
#> ── Calibration Verification ────────────────────────────────────────────────────
#> ✔ Overall: PASS (3/3 samples within 15%)
#> 
#> ── Sample Results ──
#> 
#> # A tibble: 3 × 8
#>   sample_id nominal_conc response predicted_conc accuracy_pct deviation_pct
#>   <chr>            <dbl>    <dbl>          <dbl>        <dbl>         <dbl>
#> 1 QC_Low               3      3.1           2.82         94.1         -5.93
#> 2 QC_Mid              75     77.5          73.2          97.7         -2.34
#> 3 QC_High            400    395.          374.           93.5         -6.51
#> # ℹ 2 more variables: acceptance_limit <dbl>, pass <lgl>

Limits of Detection and Quantitation

Multiple Methods

measure supports multiple approaches for calculating LOD/LOQ:

# Blank-based approach (3σ/10σ)
blank_data <- data.frame(
  sample_type = rep("blank", 10),
  response = rnorm(10, mean = 0.5, sd = 0.08)
)

lod_result <- measure_lod(
  blank_data,
  "response",
  method = "blank_sd",
  calibration = cal
)
print(lod_result)
#> <measure_lod>
#>   Value: 0.8726
#>   Method: blank_sd
#>   k: 3
#>   Uncertainty: 0.1142
#>   Parameters:
#>     blank_mean: 0.5115
#>     blank_sd: 0.1203
#>     n_blanks: 10

# Or calculate both together
lod_loq <- measure_lod_loq(
  blank_data,
  "response",
  method = "blank_sd",
  calibration = cal
)
tidy(lod_loq)
#> # A tibble: 2 × 5
#>   limit_type value method       k uncertainty
#>   <chr>      <dbl> <chr>    <dbl>       <dbl>
#> 1 LOD        0.873 blank_sd     3       0.114
#> 2 LOQ        1.72  blank_sd    10       0.381

Precision Studies

Repeatability (Within-Run Precision)

# Data from replicate measurements
repeat_data <- data.frame(
  sample_id = rep(c("Low", "Mid", "High"), each = 6),
  concentration = c(
    rnorm(6, 10, 0.5),
    rnorm(6, 100, 4),
    rnorm(6, 500, 18)
  )
)

repeatability <- measure_repeatability(
  repeat_data,
  "concentration",
  group_col = "sample_id"
)
print(repeatability)
#> measure_precision: repeatability 
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Group: Low 
#>   n = 6 
#>   Mean = 9.82 
#>   SD = 0.7645 
#>   CV = 7.8 %
#>   95% CI: [9.017, 10.62]
#> 
#> Group: Mid 
#>   n = 6 
#>   Mean = 99.51 
#>   SD = 4.893 
#>   CV = 4.9 %
#>   95% CI: [94.37, 104.6]
#> 
#> Group: High 
#>   n = 6 
#>   Mean = 501.1 
#>   SD = 18.58 
#>   CV = 3.7 %
#>   95% CI: [481.6, 520.6]

The plot keeps the six replicate results visible at each level. Diamonds mark the group means; horizontal jitter separates observations with similar values.

ggplot(
  repeat_data,
  aes(
    x = factor(sample_id, levels = c("Low", "Mid", "High")),
    y = concentration
  )
) +
  geom_point(
    position = position_jitter(width = 0.08, height = 0, seed = 42),
    colour = "#2166AC",
    size = 2
  ) +
  stat_summary(
    fun = mean,
    geom = "point",
    shape = 23,
    size = 3,
    fill = "white",
    colour = "#B2182B"
  ) +
  labs(
    x = "Sample level",
    y = "Concentration (simulated scale)"
  )
Jittered dot plot of six concentration replicates for each of three levels, ordered low, mid, high from left to right. Concentration is on the y-axis; white diamonds mark the group means. The vertical spread shows within-level variation in these simulated results.

Six simulated repeatability measurements at each of the low, mid, and high sample levels, ordered from left to right. Diamonds mark the mean within each level; point positions are jittered horizontally only to separate observations.

Intermediate Precision

# Data from multiple days
ip_data <- data.frame(
  day = rep(1:3, each = 6),
  analyst = rep(c("A", "A", "A", "B", "B", "B"), 3),
  concentration = 100 +
    rep(c(-2, 0, 2), each = 6) +  # Day effect
    rep(c(-1, 1), 9) +            # Analyst effect
    rnorm(18, sd = 3)             # Residual
)

ip_result <- measure_intermediate_precision(
  ip_data,
  "concentration",
  factors = c("day", "analyst")
)
print(ip_result)
#> measure_precision: intermediate 
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Variance Components:
#>   day: 15.78 (57%)
#>   analyst: 3.82 (14%)
#>   Residual: 8.135 (29%)
#> 
#> CV by component:
#>   day: 4%
#>   analyst: 2%
#>   Residual: 2.9%

Gage R&R Analysis

For measurement system analysis:

# Gage R&R data
grr_data <- data.frame(
  part = rep(1:5, each = 6),
  operator = rep(rep(c("Op1", "Op2"), each = 3), 5),
  measurement = c(
    # Part 1
    10.1, 10.2, 10.0, 10.3, 10.1, 10.2,
    # Part 2
    20.2, 20.1, 20.3, 20.0, 20.2, 20.1,
    # Part 3
    15.1, 15.0, 15.2, 15.3, 15.1, 15.0,
    # Part 4
    25.0, 25.1, 24.9, 25.2, 25.0, 25.1,
    # Part 5
    30.1, 30.2, 30.0, 30.1, 30.0, 30.2
  )
)

grr_result <- measure_gage_rr(
  grr_data,
  "measurement",
  part_col = "part",
  operator_col = "operator"
)
#> Warning: Negative variance components set to zero for "interaction" and "operator". This
#> may indicate insufficient variation or poor model fit.
print(grr_result)
#> measure_gage_rr: Measurement System Analysis
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Study design:
#>   Parts: 5 
#>   Operators: 2 
#>   Replicates: 3 
#> 
#> Variance Components:
#>   Repeatability: 0.01133 (0.02% contribution)
#>   Reproducibility: 0 (0% contribution)
#>   Total R&R: 0.01133 (0.02% contribution)
#>   Part-to-Part: 62.08 (100% contribution)
#> 
#> % Study Variation:
#>   Repeatability: 1%
#>   Reproducibility: 0%
#>   Total R&R: 1%
#>   Part-to-Part: 100%
#> 
#> Number of Distinct Categories (ndc): 104 
#> 
#> Assessment:
#>   Measurement system is ACCEPTABLE (%R&R < 10%)

Accuracy Assessment

Bias and Recovery

accuracy_data <- data.frame(
  level = rep(c("Low", "Mid", "High"), each = 5),
  measured = c(
    rnorm(5, 10.2, 0.3),   # Low level, slight positive bias
    rnorm(5, 100, 2.5),    # Mid level, no bias
    rnorm(5, 498, 8)       # High level, slight negative bias
  ),
  reference = rep(c(10, 100, 500), each = 5)
)

accuracy <- measure_accuracy(
  accuracy_data,
  "measured",
  "reference",
  group_col = "level"
)
print(accuracy)
#> measure_accuracy
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Group: Low 
#>   n = 5 
#>   Mean measured = 10.09 
#>   Mean reference = 10 
#>   Bias = 0.08854 ( 0.89 %)
#>   Recovery = 101 %
#>   Recovery 95% CI: [95%, 106%]
#> 
#> Group: Mid 
#>   n = 5 
#>   Mean measured = 101 
#>   Mean reference = 100 
#>   Bias = 1.042 ( 1 %)
#>   Recovery = 101 %
#>   Recovery 95% CI: [99%, 103%]
#> 
#> Group: High 
#>   n = 5 
#>   Mean measured = 502.6 
#>   Mean reference = 500 
#>   Bias = 2.593 ( 0.52 %)
#>   Recovery = 101 %
#>   Recovery 95% CI: [99%, 102%]

Linearity Assessment

linearity_data <- data.frame(
  concentration = rep(c(10, 25, 50, 75, 100), each = 3),
  response = rep(c(10, 25, 50, 75, 100), each = 3) * 1.02 +
             rnorm(15, sd = 1.5)
)

linearity <- measure_linearity(
  linearity_data,
  "concentration",
  "response"
)
print(linearity)
#> measure_linearity
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Data:
#>   n = 15 ( 5 levels )
#>   Range: 10 - 100 
#> 
#> Regression:
#>   Slope = 1.023 
#>     95% CI: [1.003, 1.044]
#>   Intercept = -0.1535 
#>     95% CI: [-1.436, 1.129]
#> 
#> Fit Quality:
#>   R-squared = 0.99884 
#>   Adj. R-squared = 0.99875 
#>   Residual SD = 1.222 
#>   Residual CV = 2.3 %
#> 
#> Lack-of-Fit Test:
#>   F = 0.649 
#>   p-value = 0.6015 
#>   Result: Not significant (linearity acceptable)

# Plot with fit line
autoplot(linearity, type = "fit") +
  labs(
    x = "Concentration (simulated scale)",
    y = "Response (simulated scale)"
  )
Scatter chart of response against concentration for 15 simulated observations. Points form five concentration groups with three observations each; a fitted straight line is overlaid. Both axes use the simulated data scale.

Fifteen simulated observations across five concentration levels, with the fitted linear model. The plot describes this generated example and does not establish linearity for a method.

Uncertainty Budgets

ISO GUM Uncertainty

Create uncertainty budgets following the GUM (Guide to the Expression of Uncertainty in Measurement):

# Define uncertainty components
components <- list(
  uncertainty_component(
    name = "Repeatability",
    type = "A",
    value = 0.5,
    df = 9
  ),
  uncertainty_component(
    name = "Calibration",
    type = "B",
    value = 0.3,
    distribution = "normal"
  ),
  uncertainty_component(
    name = "Reference Standard",
    type = "B",
    value = 0.1,
    distribution = "rectangular"
  ),
  uncertainty_component(
    name = "Temperature",
    type = "B",
    value = 0.2,
    sensitivity = 0.5  # Sensitivity coefficient
  )
)

budget <- measure_uncertainty_budget(.list = components)
print(budget)
#> <measure_uncertainty_budget>
#>   Components: 4 (1 Type A, 3 Type B)
#>   Combined u: 0.6
#>   Effective df: 19
#>   Coverage k: 2
#>   Expanded U: 1.2

Visualizing Uncertainty Contributions

autoplot(budget) +
  guides(fill = guide_legend(title = "Component type"))
Pareto chart of percentage contribution to variance for Repeatability, Calibration, Reference Standard, and Temperature, ordered from largest to smallest. Bars use fill to distinguish Type A from Type B; a line with points shows the cumulative percentage on the secondary axis.

The simulated uncertainty budget ranks four components by their percentage contribution to combined variance. Bar fill distinguishes Type A and Type B components; the line tracks the cumulative contribution.

Control Charts

Setting Up Control Limits

# Historical QC data
qc_history <- data.frame(
  run_order = 1:30,
  qc_value = rnorm(30, mean = 100, sd = 2)
)

limits <- measure_control_limits(qc_history, "qc_value")
print(limits)
#> measure_control_limits: shewhart chart
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#>   n = 30 
#>   Center = 100.1 
#>   Sigma = 1.737 
#>   UCL (+3s) = 105.3 
#>   UWL (+2s) = 103.6 
#>   LWL (-2s) = 96.61 
#>   LCL (-3s) = 94.88

Monitoring with Westgard Rules

# New run data including potential out-of-control point
new_run <- data.frame(
  run_order = 1:20,
  qc_value = c(rnorm(19, 100, 2), 108)  # Last point is high
)

chart <- measure_control_chart(
  new_run,
  "qc_value",
  "run_order",
  limits = limits,
  rules = c("1_3s", "2_2s", "R_4s", "10x")
)
print(chart)
#> measure_control_chart
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Observations: 20 
#> Rules applied: 1_3s, 2_2s, R_4s, 10x 
#> Violations detected: 4 
#> 
#> Status: OUT OF CONTROL
#> 
#> Violation summary:
#> # A tibble: 4 × 3
#>   run_order qc_value violation
#>       <int>    <dbl> <chr>    
#> 1         4    105.  1:3s R:4s
#> 2         5     97.3 R:4s     
#> 3        19    100.  R:4s     
#> 4        20    108   1:3s R:4s
autoplot(chart) +
  labs(
    x = "Run order",
    y = "QC response (simulated scale)"
  )
Control chart of QC response by run order. A green solid center line, orange dotted warning limits, and red dashed action limits run horizontally. Four points are flagged; run 20 is the highest and lies above the upper action limit.

The simulated QC run is plotted against the green solid center line, orange dotted warning limits, and red dashed action limits estimated from historical QC results. Point colour indicates whether a rule violation was flagged.

Acceptance Criteria

Defining Criteria

# Create custom criteria
my_criteria <- measure_criteria(
  criterion("cv", "<=", 15, description = "Precision CV"),
  criterion("bias_pct", "between", c(-10, 10), description = "Bias"),
  criterion("recovery", "between", c(85, 115), description = "Recovery %")
)
print(my_criteria)
#> <measure_criteria> with 3 criteria
#>   • Precision CV
#>   • Bias
#>   • Recovery %

Using Preset Criteria

# ICH Q2 presets
ich_criteria <- criteria_ich_q2()
print(ich_criteria)
#> <measure_criteria> with 4 criteria
#>   • Repeatability RSD <= 2%
#>   • Intermediate precision RSD <= 5%
#>   • Recovery 98-102%
#>   • R² >= 0.999

# Bioanalytical presets
bio_criteria <- criteria_bioanalytical()
print(bio_criteria)
#> <measure_criteria> with 5 criteria
#>   • QC CV <= 15%
#>   • Calibration CV <= 20%
#>   • R² >= 0.99
#>   • Recovery 80-120%
#>   • Bias within +/-15%

Assessing Results

# Sample results to assess (single summary values per criterion)
# For example, from a method validation summary
results <- list(
  cv = 5.2,          # Overall precision CV
  bias_pct = 1.3,    # Overall bias
  recovery = 101.3   # Mean recovery
)

assessment <- measure_assess(results, my_criteria)
print(assessment)
#> <measure_assessment> [PASS]
#>   3 passed, 0 failed
#> 
#>   ✓ cv: 5.2 (<= 15)
#>   ✓ bias_pct: 1.3 (between [-10, 10])
#>   ✓ recovery: 101.3 (between [85, 115])

# Check if all criteria passed
all_pass(assessment)
#> [1] TRUE

Method Comparison

When validating a new method, you often need to compare it against a reference or existing method. The measure package provides several approaches for method comparison studies.

Bland-Altman Analysis

Bland-Altman plots show the agreement between two methods by plotting differences against means:

# Paired measurements from two methods
comparison_data <- data.frame(
  sample_id = 1:30,
  method_A = rnorm(30, mean = 100, sd = 15),
  method_B = rnorm(30, mean = 102, sd = 16)
)

ba <- measure_bland_altman(
  comparison_data,
  method1_col = "method_A",
  method2_col = "method_B",
  regression = "linear"  # Test for proportional bias
)
print(ba)
#> measure_bland_altman
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Bias Statistics:
#>   n = 30 
#>   Mean bias = 4.083 
#>   SD of differences = 17.97 
#>   95% CI for bias: [-2.625, 10.79]
#> 
#> Limits of Agreement:
#>   Lower LOA = -31.13 (95% CI: [ -42.75 ,  -19.51 ])
#>   Upper LOA = 39.3 (95% CI: [ 27.68 ,  50.92 ])
#>   LOA Width = 70.43 
#> 
#> Proportional Bias Test:
#>   Slope = 0.2046 
#>   p-value = 0.57 
#>   Result: No significant proportional bias
autoplot(ba) +
  labs(
    x = "Mean of methods (simulated scale)",
    y = "Difference, method B minus method A (simulated scale)"
  )
#> `geom_smooth()` using formula = 'y ~ x'
Bland–Altman scatter chart for 30 simulated pairs. Mean of the two methods is on the x-axis and method B minus method A is on the y-axis. A solid line marks mean bias, dashed lines mark the estimated limits of agreement, and a dotted line shows the fitted linear trend.

Differences between simulated method B and method A results are plotted against their pairwise means. Horizontal lines mark the estimated mean bias and limits of agreement; a dotted line shows the tested linear trend. These data are illustrative only.

Regression Methods

For method comparison regression, use Deming or Passing-Bablok regression which account for error in both methods:

# Method comparison with known measurement error
deming_data <- data.frame(
  reference = c(5, 10, 25, 50, 100, 200, 400),
  test_method = c(5.2, 10.3, 25.8, 51.2, 101.5, 203.1, 408.2)
)

deming <- measure_deming_regression(
  deming_data,
  method1_col = "reference",
  method2_col = "test_method",
  bootstrap = TRUE,
  bootstrap_n = 500
)
#> Using default error ratio of 1. Provide `error_ratio` or SDs for more accurate
#> results.
print(deming)
#> measure_deming_regression
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Coefficients:
#> # A tibble: 2 × 4
#>   term      estimate ci_lower ci_upper
#>   <chr>        <dbl>    <dbl>    <dbl>
#> 1 intercept -0.00414   -0.325    0.437
#> 2 slope      1.02       1.01     1.02 
#> 
#> Statistics:
#>   n = 7 
#>   Error ratio = 1 
#>   RMSE = 0.4087 
#>   R² = 1 
#> 
#> (Fitted using mcr package)

# Check if methods are equivalent
glance(deming)
#> # A tibble: 1 × 6
#>   intercept slope intercept_ci_includes_0 slope_ci_includes_1 r_squared  rmse
#>       <dbl> <dbl> <lgl>                   <lgl>                   <dbl> <dbl>
#> 1  -0.00414  1.02 TRUE                    FALSE                   1.000 0.409

For Passing-Bablok regression (non-parametric), install the mcr package:

# Requires: install.packages("mcr")
pb <- measure_passing_bablok(
  deming_data,
  method1_col = "reference",
  method2_col = "test_method"
)
print(pb)

Proficiency Testing

Evaluate laboratory performance in proficiency testing programs:

# PT results from multiple labs
pt_data <- data.frame(
  lab_id = paste0("Lab_", 1:10),
  measured = c(99.2, 100.5, 98.8, 101.2, 97.5, 100.1, 99.8, 102.3, 100.6, 94.0),
  assigned = rep(100, 10),
  uncertainty = c(1.5, 2.0, 1.8, 1.6, 2.2, 1.9, 1.7, 2.1, 1.5, 2.0)
)

# z-scores with known sigma
z_scores <- measure_proficiency_score(
  pt_data,
  measured_col = "measured",
  reference_col = "assigned",
  score_type = "z_score",
  sigma = 2.5
)
print(z_scores)
#> measure_proficiency_score
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Score Type: z_score 
#> Sigma: 2.5 
#> 
#> Results (n = 10 ):
#>   Satisfactory (|z| ≤ 2): 9 ( 90 %)
#>   Questionable (2 < |z| ≤ 3): 1 
#>   Unsatisfactory (|z| > 3): 0 
#> 
#> Score Statistics:
#>   Mean score: -0.24 
#>   SD score: 0.925 
#>   Max |score|: 2.4
autoplot(z_scores, type = "point") +
  scale_x_continuous(breaks = seq_len(nrow(pt_data)), labels = pt_data$lab_id) +
  labs(
    x = "Laboratory",
    y = "z score",
    colour = "Score classification"
  )
Point chart of z-scores for Lab_1 through Lab_10. The y-axis is z-score; horizontal lines mark zero and plus or minus 2 and 3. Lab_10 falls below −2 and is marked questionable; the other nine points are marked satisfactory.

Proficiency z-scores for the ten simulated laboratories, using an assigned value of 100 and the specified sigma of 2.5. Horizontal reference lines mark zero and the ±2 and ±3 score boundaries; Lab_10 is classified as questionable and the other nine results as satisfactory.

Matrix Effects

Matrix effects (ion suppression/enhancement) must be evaluated in LC-MS/MS and similar methods.

Evaluating Matrix Effects

# Post-extraction spike experiment
me_data <- data.frame(
  sample_type = rep(c("matrix", "neat"), each = 6),
  matrix_lot = rep(c("Lot1", "Lot2", "Lot3"), 4),
  concentration = rep(c("low", "high"), each = 3, times = 2),
  response = c(
    # Matrix samples (some suppression)
    9200, 9500, 8900, 47500, 48200, 46800,
    # Neat samples
    10000, 10000, 10000, 50000, 50000, 50000
  )
)

me <- measure_matrix_effect(
  me_data,
  response_col = "response",
  sample_type_col = "sample_type",
  matrix_level = "matrix",
  neat_level = "neat",
  concentration_col = "concentration"
)
print(me)
#> measure_matrix_effect
#> ──────────────────────────────────────────────────────────────────────────────── 
#> 
#> Overall Matrix Effect Summary:
#>   Groups evaluated: 2 
#>   Mean ME: 94 %
#>   SD ME: 2 %
#>   CV ME: 2 %
#>   Range: 92 - 95 %
#> 
#> Effect Classification:
#>   Ion suppression (ME < 100%): 2 
#>   Ion enhancement (ME > 100%): 0 
#>   Acceptable (80-120%): 2 / 2
autoplot(me, type = "bar") +
  scale_x_discrete(labels = c("1" = "Low", "2" = "High")) +
  guides(fill = guide_legend(title = "Interpretation")) +
  labs(
    x = "Concentration group",
    y = "Matrix effect (%)",
    fill = "Interpretation"
  )
Bar chart of matrix effect percentage for low and high concentration groups. The y-axis is matrix effect percent, the solid horizontal line marks 100%, dashed lines mark 80% and 120%, and bar fill indicates the returned interpretation category.

Estimated matrix effect percentages for the low and high simulated concentration groups. The solid line marks 100% and dashed lines mark 80% and 120%; classification colours follow the result returned by measure_matrix_effect().

Standard Addition Correction

When matrix effects vary between samples, standard addition provides sample-specific correction:

library(recipes)

# Standard addition data
sa_data <- data.frame(
  sample_id = rep(c("Sample1", "Sample2"), each = 4),
  addition = rep(c(0, 10, 20, 30), 2),
  response = c(
    150, 250, 350, 450,  # Sample 1
    250, 350, 450, 550   # Sample 2
  )
)

rec <- recipe(~ ., data = sa_data) |>
  step_measure_standard_addition(
    response,
    addition_col = "addition",
    sample_id_col = "sample_id"
  ) |>
  prep()

# Original concentrations calculated via extrapolation
bake(rec, new_data = NULL)

Sample Preparation QC

Recipe steps for quality control during sample preparation.

Dilution Factor Correction

Back-calculate concentrations for diluted samples:

library(recipes)

dilution_data <- data.frame(
  sample_id = paste0("S", 1:5),
  dilution_factor = c(1, 2, 5, 10, 1),
  analyte = c(50, 45, 42, 48, 51)  # Measured after dilution
)

rec <- recipe(~ ., data = dilution_data) |>
  update_role(sample_id, new_role = "id") |>
  step_measure_dilution_correct(
    analyte,
    dilution_col = "dilution_factor",
    operation = "multiply"
  ) |>
  prep()

# Back-calculated original concentrations
bake(rec, new_data = NULL)
#> # A tibble: 5 × 3
#>   sample_id dilution_factor analyte
#>   <chr>               <dbl>   <dbl>
#> 1 S1                      1      50
#> 2 S2                      2      90
#> 3 S3                      5     210
#> 4 S4                     10     480
#> 5 S5                      1      51

Surrogate Recovery

Monitor extraction efficiency with surrogate standards:

qc_data <- data.frame(
  sample_id = paste0("QC", 1:6),
  surrogate = c(95, 105, 88, 112, 75, 132)  # Expected = 100
)

rec <- recipe(~ ., data = qc_data) |>
  update_role(sample_id, new_role = "id") |>
  step_measure_surrogate_recovery(
    surrogate,
    expected_value = 100,
    action = "flag",
    min_recovery = 80,
    max_recovery = 120
  ) |>
  prep()

# Flag samples outside recovery limits
bake(rec, new_data = NULL)
#> # A tibble: 6 × 3
#>   sample_id surrogate .surrogate_pass
#>   <chr>         <dbl> <lgl>          
#> 1 QC1              95 TRUE           
#> 2 QC2             105 TRUE           
#> 3 QC3              88 TRUE           
#> 4 QC4             112 TRUE           
#> 5 QC5              75 FALSE          
#> 6 QC6             132 FALSE

Drift Correction

Detecting Drift

# Data with drift
drift_data <- data.frame(
  sample_type = rep("qc", 20),
  run_order = 1:20,
  feature1 = 100 + (1:20) * 0.8 + rnorm(20, sd = 2),  # Has drift

  feature2 = 100 + rnorm(20, sd = 2)                   # No drift
)

drift_result <- measure_detect_drift(
  drift_data,
  features = c("feature1", "feature2"),
  qc_type = "qc"
)
print(drift_result)
#> # A tibble: 2 × 5
#>   feature    slope slope_pvalue percent_change significant
#>   <chr>      <dbl>        <dbl>          <dbl> <lgl>      
#> 1 feature1  0.630     0.0000148         11.1   TRUE       
#> 2 feature2 -0.0351    0.634             -0.662 FALSE

Correcting Drift

library(recipes)

# Using QC-LOESS correction in a recipe
rec <- recipe(~ ., data = drift_data) |>
  step_measure_drift_qc_loess(
    feature1, feature2,
    qc_type = "qc"
  ) |>
  prep()

corrected <- bake(rec, new_data = NULL)

Validation Reports

measure_validation_report() collects analysis objects, metadata, and text sections in a report object. The package includes templates organized around ICH Q2(R2) and USP <1225> sections. ### Creating a Validation Report

# Gather validation results (using objects from above)
report <- measure_validation_report(
  # Metadata
  title = "Illustrative Method Validation Report",
  method_name = "Simulated Assay",
  method_description = "Example metadata; study measurements in this vignette are simulated",
  analyst = "Example analyst",
  reviewer = "Example reviewer",
  lab = "Example laboratory",
  instrument = "Example instrument",


  # Validation sections (results from earlier in this vignette)
  calibration = cal,
  lod_loq = lod_loq,
  accuracy = accuracy,
  precision = list(repeatability = repeatability, intermediate = ip_result),
  linearity = linearity,
  range = list(lower = 1, upper = 500, units = "simulated scale"),
  uncertainty = budget,

  # Text sections
  specificity = "Example placeholder; blank-matrix results are not included in this vignette.",
  robustness = list(
    factors = c("Flow rate (±0.1 mL/min)", "Column temperature (±5°C)", "Mobile phase pH (±0.2)"),
    conclusion = "Example placeholder; robustness-study data are not included in this vignette."
  ),

  # Conclusions
  conclusions = list(
    summary = "No method suitability conclusion is supported by the simulated data in this vignette.",
    recommendations = c(
      "Repeat the analyses with representative study data for the intended use",
      "Document the acceptance criteria and any changes to instruments or reagents"
    )
  ),

  # References
  references = c(
    "ICH Q2(R2) Validation of Analytical Procedures (2023)",
    "USP <1225> Validation of Compendial Procedures"
  )
)

print(report)
#> 
#> ── Validation Report ───────────────────────────────────────────────────────────
#> Title: Illustrative Method Validation Report
#> Method: Simulated Assay
#> Analyst: Example analyst
#> Lab: Example laboratory
#> Date: 2026-11-16
#> 
#> ── Validation Sections ──
#> 
#> ℹ Calibration
#> ℹ LOD/LOQ
#> ℹ Accuracy
#> ℹ Precision
#> ℹ Linearity
#> ℹ Range
#> ℹ Specificity/Selectivity
#> ℹ Robustness
#> ℹ Measurement Uncertainty
#> 
#> ── Conclusions ──
#> 
#> No method suitability conclusion is supported by the simulated data in this
#> vignette.
#> 
#> ── Provenance ──
#> 
#> Generated: 2026-11-16 15:21:32.000916
#> R version: 4.6.2
#> measure version: 0.0.1.9003
#> 
#> ℹ Use `render_validation_report()` to generate document

Inspecting Report Sections

# Check which sections are included
summary(report)
#> 
#> ── Validation Report Summary ───────────────────────────────────────────────────
#> Method: Simulated Assay
#> Date: 2026-11-16
#> 
#> # A tibble: 9 × 4
#>   section                 status n_results notes             
#>   <chr>                   <chr>      <int> <chr>             
#> 1 Calibration             info           8 ""                
#> 2 LOD/LOQ                 info          NA "Method: blank_sd"
#> 3 Accuracy                info           3 ""                
#> 4 Precision               info          NA ""                
#> 5 Linearity               info          NA ""                
#> 6 Range                   info          NA ""                
#> 7 Specificity/Selectivity info          NA ""                
#> 8 Robustness              info          NA ""                
#> 9 Measurement Uncertainty info          NA ""
#> 
#> ✔ All sections meet acceptance criteria

# Access specific sections
has_validation_section(report, "calibration")
#> [1] TRUE
has_validation_section(report, "stability")  # Not included
#> [1] FALSE

# Get section data
get_validation_section(report, "range")
#> $lower
#> [1] 1
#> 
#> $upper
#> [1] 500
#> 
#> $units
#> [1] "simulated scale"

Adding Custom Sections

You can add additional sections after report creation:

# Add a stability section later
report <- add_validation_section(
  report,
  "stability",
  list(
    description = "Short-term stability at room temperature",
    results = data.frame(
      timepoint = c("0h", "4h", "8h", "24h"),
      recovery_pct = c(100, 99.5, 98.8, 97.2)
    ),
    conclusion = "Sample is stable for 24 hours at room temperature."
  )
)

has_validation_section(report, "stability")
#> [1] TRUE

Tidy Output

Extract all results as a tidy tibble for further analysis:

tidy(report)
#> # A tibble: 10 × 41
#>    section              term  estimate std_error statistic   p_value group     n
#>    <chr>                <chr>    <dbl>     <dbl>     <dbl>     <dbl> <chr> <int>
#>  1 Calibration          (Int…    0.119    0.0395      3.00  2.39e- 2 NA       NA
#>  2 Calibration          nomi…    1.06     0.0143     73.8   4.16e-10 NA       NA
#>  3 Accuracy             NA      NA       NA          NA    NA        Low       5
#>  4 Accuracy             NA      NA       NA          NA    NA        Mid       5
#>  5 Accuracy             NA      NA       NA          NA    NA        High      5
#>  6 Linearity            NA      NA       NA          NA    NA        NA       15
#>  7 Measurement Uncerta… NA      NA       NA          NA    NA        NA       NA
#>  8 Measurement Uncerta… NA      NA       NA          NA    NA        NA       NA
#>  9 Measurement Uncerta… NA      NA       NA          NA    NA        NA       NA
#> 10 Measurement Uncerta… NA      NA       NA          NA    NA        NA       NA
#> # ℹ 33 more variables: mean_measured <dbl>, mean_reference <dbl>, bias <dbl>,
#> #   bias_pct <dbl>, mean_recovery <dbl>, sd_recovery <dbl>, cv_recovery <dbl>,
#> #   recovery_ci_lower <dbl>, recovery_ci_upper <dbl>, n_levels <int>,
#> #   range_min <dbl>, range_max <dbl>, slope <dbl>, slope_ci_lower <dbl>,
#> #   slope_ci_upper <dbl>, intercept <dbl>, intercept_ci_lower <dbl>,
#> #   intercept_ci_upper <dbl>, r_squared <dbl>, adj_r_squared <dbl>,
#> #   residual_sd <dbl>, residual_cv <dbl>, lof_f <dbl>, lof_p <dbl>, …

Rendering Reports

To render a validation report to HTML or PDF, use render_validation_report(). Two templates are provided:

  • ICH Q2(R2): Organized by validation characteristics (specificity, linearity, range, accuracy, precision, etc.)
  • USP <1225>: Organized by procedure category (I, II, III, IV)
# Render to HTML using ICH Q2 template (default)
render_validation_report(
  report,
  output_file = "validation_report.html",
  template = "ich_q2"
)

# Render to PDF using USP <1225> template
render_validation_report(
  report,
  output_file = "validation_report.pdf",
  output_format = "pdf",
  template = "usp_1225"
)

The rendered report includes:

  • Header: Method name, date, analyst, reviewer, lab, instrument
  • Table of Contents: Auto-generated from sections
  • Validation Sections: Each with formatted tables and plots
  • Provenance: R version, package versions, timestamp for reproducibility

Summary

Functions shown in this vignette include:

Category Key Functions
Calibration measure_calibration_fit(), measure_calibration_predict(), measure_calibration_verify()
LOD/LOQ measure_lod(), measure_loq(), measure_lod_loq()
Precision measure_repeatability(), measure_intermediate_precision(), measure_gage_rr()
Accuracy measure_accuracy(), measure_linearity(), measure_carryover()
Method Comparison measure_bland_altman(), measure_deming_regression(), measure_passing_bablok(), measure_proficiency_score()
Matrix Effects measure_matrix_effect(), step_measure_standard_addition()
Sample Prep QC step_measure_dilution_correct(), step_measure_surrogate_recovery()
Uncertainty measure_uncertainty_budget(), measure_uncertainty()
Control Charts measure_control_limits(), measure_control_chart()
Criteria measure_criteria(), measure_assess(), criteria_ich_q2(), criteria_bland_altman(), criteria_matrix_effects()
Drift measure_detect_drift(), step_measure_drift_qc_loess()
Validation Reports measure_validation_report(), render_validation_report()

The examples use simulated measurements and illustrative report text. They demonstrate API results, not evidence that a method meets criteria or is suitable for a particular intended use. Check each function’s help page for its inputs and assumptions.

See Also