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.

After following the instructions to reproduce the analysis, you will also need to change the output_function column to tell the pf_output function which check you executed. Reference the table below for the labels that are associated with each check:

Check Type	output_function
Multi Site, Exploratory, Cross-Sectional	pf_ms_exp_cs
Multi Site, Exploratory, Longitudinal	pf_ms_exp_la
Multi Site, Anomaly Detection, Cross-Sectional	pf_ms_anom_cs
Multi Site, Anomaly Detection, Longitudinal	pf_ms_anom_la

Multi-Site Exploratory Analysis

Cross-Sectional

First, execute the Single Site, Exploratory, Cross-Sectional analysis, configured appropriately for your study, against each data source.

library(patientfacts)

my_table <- pf_process(cohort = my_cohort,
                       multi_or_single_site = 'single',
                       anomaly_or_exploratory = 'exploratory',
                       time = F,
                       patient_level_tbl = T / F,
                       ...)

The median_all_* columns will not reflect the overall median using this method. This is typically used as a baseline comparison metric in pf_output. If you do not mind the lack of an overall comparison metric and would just like to compare site-level medians, all you need to do is union each table together and adjust the output_function accordingly.

my_final_results <- my_table1 %>% dplyr::union(my_table2) ... %>%
  dplyr::union(my_table_n) %>%
  dplyr::mutate(output_function = 'pf_ms_exp_cs')

If you would like to have the overall median available, you will need to output patient-level results in addition to the standard summary output (patient_level_tbl = T). Once patient-level output is returned from each data source, you will need to union these tables together and pass them through the functions below.

my_combined_results <- my_table1 %>% dplyr::union(my_table2) ... %>%
  dplyr::union(my_table_n) %>%
  dplyr::mutate(output_function = 'pf_ms_exp_cs')

my_final_results <- 
  patientfacts:::compute_pf_medians(data_input = my_combined_results,
                                    site_col = 'site',
                                    agegrp = age_groups # (as provided in pf_process)
                                    )

Longitudinal

The Longitudinal analysis does not output any overall comparative median, so for this check, just execute the Single Site, Exploratory, Longitudinal check against each data source and union the results together.

Outputting patient level results is not required, but this option is still available for the longitudinal analysis if needed.

my_table <- pf_process(cohort = my_cohort,
                       multi_or_single_site = 'single',
                       anomaly_or_exploratory = 'exploratory',
                       time = T,
                       patient_level_tbl = T / F,
                       ...)

my_final_results <- my_table1 %>% dplyr::union(my_table2) ... %>%
  dplyr::union(my_table_n) %>%
  dplyr::mutate(output_function = 'pf_ms_exp_la')

Multi-Site Anomaly Detection Analysis

Cross-Sectional

Follow the instructions above for producing the Multi-Site, Exploratory, Cross-Sectional output with patient-level results. This is required for this anomaly detection process as it is not possible to back-compute from the standard aggregated results.

Once a combined table of patient-level results from each data source has been created, execute the code below (using functions from the squba.gen package) to run the anomaly detection analysis. The p_value can be selected by the user.

my_summary_table <- my_combo_results %>% 
  group_by(site, visit_type, domain) %>%
  summarise(tot_pt = n_distinct(person_id), 
            n_pt_fact = sum(var_ever)) %>%
  mutate(prop_pt_fact = n_pt_fact / tot_pt)

df_start <- compute_dist_anomalies(df_tbl = my_summary_table,
                                   grp_vars = c('domain', 'visit_type'),
                                   var_col = 'prop_pt_fact',
                                   denom_cols = c('domain', 'visit_type', 
                                                  'tot_pt'))

df_final <- detect_outliers(df_tbl = df_start,
                            tail_input = 'both',
                            p_input = p_value,
                            column_analysis = 'prop_pt_fact',
                            column_variable = c('domain', 'visit_type'))

Longitudinal

For a longitudinal analysis, start by following the instructions for the Multi-Site, Exploratory, Longitudinal analysis. Then, the ms_anom_euclidean function, available through the squba.gen package, should be executed against your results. Copy the code below, inputting the data you generated.

df_final <- ms_anom_euclidean(fot_input_tbl = my_table %>% 
                                mutate(prop_pts_fact = pts_w_fact / pts_w_visit),
                              grp_vars = c('site', 'visit_type', 'domain'),
                              var_col = 'prop_pts_fact')