Multi-Site Analysis for Independent Data Sources

The multi-site analyses included in this suite are intended to be executed against data that are all stored in the same place. However, there may be some instances where the data associated with each site is stored in independent locations. This vignette outlines how the multi-site analysis can be executed in these instances.

Multi-Site Exploratory Analysis

First, execute the Single Site, Exploratory analyses, configured appropriately for your study, against each data source.

library(sensitivityselectioncriteria)

my_table <- ssc_process(base_cohort = my_cohort,
                        alt_cohorts = list('My Alternate Cohort' = my_alt_cohort),
                        omop_or_pcornet = 'omop',
                        multi_or_single_site = 'single',
                        anomaly_or_exploratory = 'exploratory',
                        ...)

This function will produce 2 tables. Select the first table in the list from each result set, then combine these results into a single table with the different sites delineated in the site column. You will also need to edit the output_function column to reflect that this table now should be considered a Multi Site, Exploratory output.

my_final_results <- my_table1[[1]] %>% dplyr::union(my_table2[[1]]) ... %>%
  dplyr::union(my_table_n[[1]]) %>%
  dplyr::mutate(output_function = 'ssc_ms_exp_cs')

Multi-Site Anomaly Detection Analysis

It is slightly more complex to reproduce the anomaly detection analysis in this case, due to the level of summarization that is output by the typical function. Instead of running the ssc_process as you normally would, you will need to use an internal function to produce the correct results.

The first step will be to run the appropriate internal function, depending on which CDM you are using, against each of the data sources. It intakes most of the same parameters as the primary ssc_process function.

## For an OMOP CDM:
my_omop_table <- 
  sensitivityselectioncriteria:::compare_cohort_def_omop(
                                    base_cohort = my_cohort,
                                    alt_cohorts = list('My Alternate Cohort' = my_alt_cohort),
                                    multi_or_single_site = 'single',
                                    ...)

## For a PCORnet CDM:
my_pcornet_table <- 
  sensitivityselectioncriteria:::compare_cohort_def_pcnt(
                                    base_cohort = my_cohort,
                                    alt_cohorts = list('My Alternate Cohort' = my_alt_cohort),
                                    multi_or_single_site = 'single',
                                    ...)

Once this function has been executed against each data source, combine these results into a single table with the different sites delineated in the site column.

my_combo_results <- my_table1 %>% dplyr::union(my_table2) ... %>%
  dplyr::union(my_table_n)

Then, pass this combined table through the anomaly detection function that will execute a standardized mean difference computation. If you provided custom demographic mappings, extract the demographic labels from this file. Otherwise, use the appropriate package-provided table based on your data model (either ssc_omop_demographics or ssc_pcornet_demographics).

You will also need to edit the output_function column to reflect that this table now should be considered a Multi Site, Anomaly Detection output.

my_final_results <- 
  sensitivityselectioncriteria:::compare_cohort_smd(cohort_def_output = my_combo_results,
                                                    demographic_vector = 
                                                      my_demographic_table %>%
                                                        select(demographic) %>% pull()) %>%
  dplyr::mutate(output_function = 'ssc_ms_anom_cs')