Skip to contents

ADNIMERGE2-PACC-SCORE

Last Updated: October 08, 2025

Source: vignettes/ADNIMERGE2-PACC.Rmd
ADNIMERGE2-PACC.Rmd

Introduction

This article describes the procedure to calculate a modified version of the Preclinical Alzheimer Cognitive Composite PACC score in ADNI study based on Donohue et al. (2014) approach.

According to Donohue et al. (2014), the PACC score is a baseline standardize composite of:

  • Free and Cued Selective Reminding Test (FCSRT) score

  • Logical Memory IIa Delayed Recall (LM) score

  • Digit Symbol Substitution Test (DSST) score, and

  • Mini-Mental State Examination (MMSE) total score

Each of component score is required to be standardized by the corresponding baseline score prior to computing the composite score. Higher values of the composite score indicate better performance.

In ADNI study, some of these components are either collected partially during certain study phase or not available at all. Since FSCRT is not available at all, the Delayed Recall portion of the Alzheimer’s Disease Assessment Scale (ADAS), which is called ADAS-Cog Q4 score, will be used as a proxy. The ADAS-Cog Q4 score ranges from 0 to 10 points where higher value indicates worse condition. Similarly, a logarithm scale of Trails B score will be used as a proxy for DSST. Therefore, there will have two modified version of PACC composite score:

  1. mPACCdigit: It will only be computed in ADNI1 study phase since DSST was collected in the early phase of ADNI study.

  2. mPACCtrailsB: It will be computed across all ADNI study phases using Trails B score as a proxy for DSST.

Load Required R Packages 

# ADNI study R data package
library(ADNIMERGE2)

Compute PACC Composite Score

Import Raw Datasets

First, we import pre-processed raw datasets that contain PACC component from the ADNIMERGE2 package. These raw datasets were stored in csv format as external data in the package. The reasons for sourcing those dataset is 1) not all PACC component score are yet available at the data-sharing platform, particularly the ADAS-Cog Q4 score, and 2) to reduced any potential additional package dependency.

In the next r chunk, all raw datasets will be loaded to the current working environment using load_csv_files() function.

# Import PACC input raw datasets
pacc_raw_data_dir <- system.file(
  "extradata/pacc-raw-input",
  package = "ADNIMERGE2",
  mustWork = TRUE
)
load_csv_files(input_dir = pacc_raw_data_dir)

These raw datasets are in a long format and presented as follow:

  • pacc_mmse_long: A dataset contains the MMSE score downloaded from the data-sharing platform.

  • pacc_neurobat: A dataset contains Logical Memory - Delayed Recall (LDELTOTL), Digit Symbol Substitution (DIGITSCR), and Trials B (TRABSCOR) raw scores from Neuropsychological Battery (NEUROBAT) data. The NEUROBAT() data is also downloaded from the data-sharing platform.

  • pacc_adas_q4score_long: A dataset contains the raw ADAS-Cog Q4 score that obtained from two sources:

    • Current study phase EDC portal

    • Old version of ADNI study R data package ADNIMERGE, which is also available at the data-sharing platform, that contain records from ADNI1 to ADNI3 study phases.

Data Preparation Assumption

  • During the raw dataset preparation, the screening fail visit code in ADNI1 study phase was remapped to the actual screening visit code.

  • Since some of the PACC component assessments were either not collected at baseline visit during the whole study or partially collected at baseline visit in certain study phase. It is reasonable to carrying forward the screening visit record of an assessment as a corresponding baseline record if the baseline assessment result is missing/unknown. For instance, MMSE was not collected at any baseline visit, and the other neuropsychological assessments (i.e., LDELTOTL, DIGITSCR, and TRABSCOR) are partially collected at baseline visit in certain study phase.

The following r chuck demonstrate how the screening-baseline visit adjustment is preformed by carrying forward any screening visit record as a baseline record using adjust_scbl_record() function.

# Adjustment of screening-baseline record
id_vars <- c("COLPROT", "RID", "VISCODE", "SCORE_SOURCE")

## MMSE ----
pacc_mmse_long <- pacc_mmse_long %>%
  bind_rows(
    adjust_scbl_record(
      .data = pacc_mmse_long,
      adjust_date_col = "VISDATE",
      check_col = "SCORE"
    ) %>%
      filter(VISCODE %in% get_baseline_vistcode())
  ) %>%
  assert_uniq(all_of(id_vars)) %>%
  set_as_tibble()

## NEUROBAT ----
neurobat_vars <- c("LDELTOTL", "DIGITSCR", "TRABSCOR")

adjusted_bl_neurobat <- adjust_scbl_record(
  .data = pacc_neurobat_long,
  wide_format = FALSE,
  extra_id_cols = "SCORE_SOURCE",
  adjust_date_col = "VISDATE",
  check_col = "SCORE"
) %>%
  # Only baseline records
  filter(VISCODE %in% get_baseline_vistcode())

pacc_neurobat_long <- pacc_neurobat_long %>%
  filter(!VISCODE %in% get_baseline_vistcode()) %>%
  bind_rows(adjusted_bl_neurobat) %>%
  distinct() %>%
  assert_uniq(all_of(id_vars)) %>%
  set_as_tibble()

# ADAS-cog Q4 Score ----
pacc_adas_q4score_long <- pacc_adas_q4score_long %>%
  set_as_tibble()

The next step is to combine all the raw datasets and converting the combined data into a wide format using the study data collection phase/protocol and study visit code per subject.

# Bind PACC input raw-data
pacc_input_data_long <- bind_rows(
  pacc_mmse_long,
  pacc_neurobat_long,
  pacc_adas_q4score_long
)

# Convert into wide format
pacc_component <- c("ADASQ4SCORE", "MMSE", neurobat_vars)

pacc_input_data_wide <- pacc_input_data_long %>%
  unite(col = "SCORE_VISDATE", SCORE, VISDATE, sep = "-/") %>%
  pivot_wider(
    names_from = SCORE_SOURCE,
    values_from = SCORE_VISDATE
  ) %>%
  assert_uniq(all_of(c("COLPROT", "RID", "VISCODE"))) %>%
  mutate(across(all_of(pacc_component), ~ strsplit(as.character(.), "-/"))) %>%
  unnest_wider(all_of(pacc_component), names_sep = "_") %>%
  rename_with(~ str_remove_all(.x, "\\_1$"), starts_with(pacc_component)) %>%
  rename_with(~ str_replace_all(.x, "2$", "VISDATE"), starts_with(pacc_component)) %>%
  mutate(across(starts_with(pacc_component), ~ ifelse(.x %in% "NA", NA, as.character(.x)))) %>%
  mutate(across(all_of(pacc_component), as.numeric))

Some of the PACC component might not be collected on the same date although those assessments could be collected at the same study visit. Additionally, some of component specific screening records were carried forward as a baseline record including screening visit/assessment collection date. In order to have a representative common date associated with the PACC score per study visit, we compare the specific study visit date with corresponding component collection date.

We use get_vars_common_date() function to get a common date per specific study visit per subject as shown in the following two r chunks. A maximum date will be selected if any of the component specific dates are different. A maximum date selection is used due to the fact that some of the screening records were carried forward as a baseline record.

# Add visit date
pacc_input_data_wide <- pacc_input_data_wide %>%
  left_join(
    REGISTRY %>%
      select(COLPROT, RID, VISCODE, EXAMDATE) %>%
      set_as_tibble(),
    by = c("COLPROT", "RID", "VISCODE")
  ) %>%
  verify(nrow(.) == nrow(pacc_input_data_wide))
# Get common date per record
pacc_input_data_wide <- get_vars_common_date(
  .data = pacc_input_data_wide,
  date_cols = paste0(pacc_component, "_VISDATE"),
  compared_ref_date = TRUE,
  ref_date_col = "EXAMDATE",
  select_method = "max",
  preferred_date_col = "COMMON_DATE"
) %>%
  rename("PACC_VISDATE" = FINAL_DATE)
pacc_input_data_wide <- pacc_input_data_wide %>%
  # Add enrollment date
  left_join(
    get_adni_enrollment(.registry = REGISTRY) %>%
      mutate(RID = as.character(RID)) %>%
      select(ORIGPROT, RID, ENRLDT = EXAMDATE, ENRLFG),
    by = "RID"
  ) %>%
  # Add baseline diagnostics status
  left_join(
    get_adni_blscreen_dxsum(
      .dxsum = DXSUM,
      phase = "Overall",
      visit_type = "baseline"
    ) %>%
      mutate(RID = as.character(RID)) %>%
      select(RID, DX = DIAGNOSIS, DXDATE = EXAMDATE) %>%
      assert_uniq(RID),
    by = "RID"
  )

Component Baseline Score Summaries

The next step is to calculate component specific baseline summaries that will be used to standardize component score. This baseline summary is based on user-defined standardization population. Here, we define the standardization population as follows:

  • New enrolled subject in ADNI study

  • Subjects with baseline Cognitive Normal (CN) diagnostics status

The following r chunk shows how the PACC component baseline score is summarized across the baseline diagnostics status based on the standardization population defined in the above. The summary result is generated using compute_score_summary() function.

NOTE: It is recommended to apply a logarithm transformation of Trials B score if the score is not already in a logarithm scale.

pacc_component_bl <- pacc_input_data_wide %>%
  mutate(TRABSCOR = log(TRABSCOR + 1)) %>%
  # Baseline visit for new enrollee
  filter(ENRLFG %in% "Yes") %>%
  group_by(RID) %>%
  filter(ORIGPROT == COLPROT) %>%
  filter(VISCODE %in% get_baseline_vistcode()) %>%
  ungroup() %>%
  assert_uniq(RID)

pacc_component_bl.summary <- compute_score_summary(
  .data = pacc_component_bl,
  wideFormat = TRUE,
  scoreVar = pacc_component,
  groupVar = "DX",
  filterGroup = "CN"
)

pacc_component_bl.summary %>%
  mutate(VAR = ifelse(VAR == "TRABSCOR", "LOG.TRABSCOR", VAR)) %>%
  # rename("Baseline DX" = DX, "Component" = VAR) %>%
  datatable(
    colnames = c("Baseline DX" = 2, "Component" = 3),
    caption = "PACC Compoonent Standardization Baseline Summary"
  )

Compute PACC Score

The composite PACC score is the sum of the standardized component score adjusted by the number of components. The following two conditions should be checked before computing the PACC score.

  • Trials B Measurement Scale: To identify whether the score is in a logarithm scale. For instance, here the Trials B score is not in a logarithm scale. We need either to map the raw score values to logarithm scale prior to using a scoring function or to specify an argument that allows to perform a logarithm transformation inside a scoring function.

  • Standardized Component Directional Effect: To verify the same directional effect across standardized component score, and reoriented standardized component score since the composite score will be the sum of the standardized scores. In ADNI study, ADAS-cog Q4 and Trials B scores have opposite directional effect compared to the remaining PACC component scores. Higher values of those two components indicates worse condition. It is required to convert the direction effect of those two standardized components. Therefore, higher values of the composite score indicates better performance.

Moreover, the composite score will only be calculated if there are at least two non-missing component scores at a given study visit per subject.

Then, we use compute_pacc_score() function to compute the two modified composite PACC scores in ADNI study.

The next r chunk shows how to calculate the PACC composite score based on a wide format input component score, a standardization baseline summary, and with a logarithm transformation of Trials B score inside in the function.

PACC <- compute_pacc_score(
  .data = pacc_input_data_wide,
  bl.summary = pacc_component_bl.summary,
  componentVars = pacc_component,
  rescale_trialsB = TRUE,
  keepComponents = TRUE,
  wideFormat = TRUE
) %>%
  mutate(RID = as.numeric(RID)) %>%
  rename_with(~ str_replace_all(.x, "\\.zscore", "Z"), ends_with(".zscore")) %>%
  create_orig_protocol()

PACC <- PACC %>%
  select(all_of(pacc_data_dic$FLDNAME)) %>%
  labelled::set_variable_labels(
    .labels = pacc_data_dic$LABEL,
    .strict = TRUE
  )
Show PACC Variables

Baseline Modified-PACC Score

The following r chunks show how to get baseline PACC composite score based on the PACC dataset. Similar result can be obtained using ADSL or ADQS datasets. For more information on how to use those datasets, please refer to ADNI-Enrollment and ADNIMERGE2-Analysis-Data articles.

# Modified PACC using DSS score
baseline_mPACCdigit <- PACC %>%
  # Only in ADNI1 study phase
  filter(COLPROT == adni_phase()[1]) %>%
  filter(ORIGPROT == COLPROT) %>%
  filter(VISCODE %in% get_baseline_vistcode()) %>%
  # Enrolled subjects
  filter(ENRLFG %in% "Yes") %>%
  assert_uniq(RID) %>%
  select(RID, ORIGPROT, mPACCdigit, DX)

# Modified PACC using Trials B score as a proxy for DSS
baseline_mPACCtrailsB <- PACC %>%
  filter(ORIGPROT == COLPROT) %>%
  filter(VISCODE %in% get_baseline_vistcode()) %>%
  # Enrolled subjects
  filter(ENRLFG %in% "Yes") %>%
  assert_uniq(RID) %>%
  select(RID, ORIGPROT, mPACCtrailsB, DX)
baseline_PACC <- baseline_mPACCtrailsB %>%
  mutate(
    SCORE = mPACCtrailsB,
    SCORE_NAME = "mPACCtrailsB"
  ) %>%
  bind_rows(
    baseline_mPACCdigit %>%
      mutate(
        SCORE = mPACCdigit,
        SCORE_NAME = "mPACCdigit"
      )
  ) %>% 
  group_by(SCORE_NAME) %>% 
  mutate(SCORE_NAME = paste0(SCORE_NAME, " (n = ", sum(!is.na(SCORE)), ")")) %>% 
  ungroup()

baseline_PACC %>%
  ggplot2::ggplot(aes(x = SCORE)) +
  geom_histogram() +
  facet_wrap(~SCORE_NAME, scales = "free_y") +
  labs(x = "Baseline Modified PACC Score", y = "Count") +
  theme_bw(base_size = 12)

Baseline Modiffied PACC score

References

Donohue, Michael C, Reisa A Sperling, Ronald Petersen, Chung-Kai Sun, Michael W Weiner, Paul S Aisen, Alzheimer’s Disease Neuroimaging Initiative, et al. 2017. “Association Between Elevated Brain Amyloid and Subsequent Cognitive Decline Among Cognitively Normal Persons.” Jama 317 (22): 2305–16.
Donohue, Michael C, Reisa A Sperling, David P Salmon, Dorene M Rentz, Rema Raman, Ronald G Thomas, Michael Weiner, Paul S Aisen, et al. 2014. “The Preclinical Alzheimer Cognitive Composite: Measuring Amyloid-Related Decline.” JAMA Neurology 71 (8): 961–70.