Shifts in ophthalmic care utilization during the COVID-19 pandemic in the United States

Charles Li, Flora Lum, Evan M. Chen, Philip A. Collender, Jennifer R. Head, Rahul N. Khurana, Emmett T. Cunningham Jr., Ramana S. Moorthy, David W. Parke II, Stephen D. McLeod

Last updated: December 17, 2023 by Charles Li (cli@aao.org), including data up to December 31, 2021

Li, C., Lum, F., Chen, E.M. et al. Shifts in ophthalmic care utilization during the COVID-19 pandemic in the US. Commun Med 3, 181 (2023). https://doi.org/10.1038/s43856-023-00416-4

About

This repository hosts the code, outputs, and supplementary information for a study focused on understanding the disruptions to ophthalmic care in the United States for the first two years of the COVID-19 pandemic. Using data from the American Academy of Ophthalmology IRIS Registry® (Intelligent Research in Sight), this research characterized patterns of eye care utilization across 261 different ocular conditions from January 2020 to December 2021 by leveraging a common analytical framework to explore factors that may explain the differential underutilization of care.

In the acute phase of the pandemic, redistributions of healthcare resources were required to minimize mortality and limit the spread of COVID-19. However, the patterns of, and reasons explaining, sustained utilization reductions in the post-acute phase are not entirely understood. Existing research by health economists have estimated the "elasticity" (i.e., responsiveness) of demand of healthcare services in response to changes in another variable (e.g., cost, income); similarly, we explored how utilization levels for a wide range of ocular diagnoses exhibited varying degrees of sensitivity to possible pandemic-related restrictions to the seeking or delivery of care (e.g., resource constraints, behavioral modifications). We specifically examined whether features of ocular diagnoses themselves—namely, disease severity—could explain shifts in care utilization patterns that were observed during the pandemic.

Data Processing and Analytic Steps

This project consists of the following main stages:

Step 1A. Construct an inventory of ophthalmic diagnoses to study

1. We first specified an inventory of conditions ("diagnosis entities") to include for analysis using the spreadsheet file codebooks/source materials/CCSR ICD10 5.4.20_modified.xlsx, adapted from v2020.2 of the U.S. Agency for Healthcare Research and Quality Clinicial Classifications Software Refined (CCSR) database. The CCSR aggregates tens of thousands of International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis codes into clinically meaningful groupings, which are called "diagnosis entities" in this study. We mainly considered codes from the "EYE" chapter of the CCSR, which encompasses ICD-10-CM codes related to diseases of the eye and adnexa, for this study.

   CCSR ICD10 5.4.20_modified.xlsx contains two tabs:

   • The first tab is a spreadsheet that maps each ICD-10-CM diagnosis code (ICD-10 Code) to a diagnosis entity (DiagnosisEntity). Each distinct ICD-10-CM code is listed once (on a single row), with the ICD-10-CM description of the diagnosis code (Diagnosis), the diagnosis entity that the ICD-10-CM code belongs to (DiagnosisEntity), and the broader diagnosis category that the diagnosis entity belongs to (CCSR Category), in abbreviation form (EYE0XX), listed across the columns of the spreadsheet. Unlike the original CCSR mapping, which allowed for some ICD-10-CM codes to be cross-classified into more than one category, we adopted a mutually exclusive categorization scheme by assigning each diagnosis entity to one of 13 diagnosis categories (EYE001 - EYE013). Furthermore, an ICD-10-CM diagnosis code with incomplete time series (TS) data of monthly counts of patients observed with that condition and/or very low monthly case counts was excluded from consideration in this study if it was not feasible to assign the code into an existing or new (standalone) diagnosis entity in a clinically meaningful way. For ICD-10-CM codes that are not assigned to any diagnosis entity, CODE EXCLUDED was written in the DiagnosisEntity column. Supplementary Note 4 of the Supplementary Information contains further details on the assignment of ICD-10-CM codes to diagnosis entities and other adaptations made from the original groupings provided by the CCSR database.

   • The second tab includes a list the full names of the diagnosis categories and their abbreviations (EYE0XX).

   For ease of import into R, both sheets of the Excel file were separately converted into CSV format:

   Outputs:
   codebooks/source materials/CCSR ICD10 dx_entities 5.4.20_modified.csv
codebooks/source materials/CCSR ICD10 categories 5.4.20_modified.csv

2. Next, we ran the R script codebooks/source materials/ccsr_codebook_qc_wrangling.Rmd to produce a clean mapping of ICD-10-CM codes and diagnosis entities, which can be subsequently imported into a relational database as a lookup table to allow monthly case numbers to be queried for each diagnosis entity. Data pre-processing steps and quality control checks were applied to remove ICD-10-CM codes designated for exclusion, format ICD-10-CM codes to ensure compatability with database queries, and verify that each diagnosis entity is assigned to exactly one diagnosis category.

   Inputs:
   codebooks/source materials/CCSR ICD10 dx_entities 5.4.20_modified.csv
codebooks/source materials/CCSR ICD10 categories 5.4.20_modified.csv
   Output:
   codebooks/ccsr_codebook_for_sql.csv

Step 1B. Extract monthly counts of patients observed with each diagnosis

Monthly numbers of patients documented with each diagnosis entity were queried from the IRIS Registry (Amazon Redshift version 1.0.38698, PostgreSQL 8.0.2). The SQL script data-extraction/covid19_alldx_pull.sql, as well as the Methods section, details the steps undertaken for data extraction, including the criteria we applied to identify diagnosis records and patients eligible for inclusion in this analysis. The final output of this script is a single table that contains the numbers of patients documented with each diagnosis entity for each month of the study period (January 2017 to Decemeber 2021).

Input:
codebooks/ccsr_codebook_for_sql.csv
Output:
data-extraction/covid_elasticity_all_dx_2017_2021_complete_20220519.csv

Note: Direct access to the IRIS Registry database is needed to run the covid19_alldx_pull.sql script. At this time, the IRIS Registry is not a publicly available dataset, but eligible investigators may apply for research opportunities to work with IRIS Registry data.

Steps 2-4. Statistical modeling & data analysis

1. After exporting a single table containing monthly time series data of case counts for each diagnosis entity, some data pre-processing and wrangling steps were undertaken via the R script data-extraction/covid_elasticity_dataset_prep.Rmd to generate a single, clean dataset for further analysis. These steps include applying appropriate formatting to year-month date information, applying appropriate string formatting to ensure consistent representations of diagnosis entity names, and generating harmonic terms with 3, 6, and 12-month periodicities.

   Input:
   data-extraction/covid_elasticity_all_dx_2017_2021_complete_20220519.csv
   Output:
   data-extraction/dx_proportions_and_cnts.csv

2. Finally, all steps undertaken for statistical modeling and data analysis (as described in Steps 2-4 in the summary graphic above) were implemented via the R script modeling-analysis/covid-elasticity-static-figures.Rmd.

   Inputs:

   • data-extraction/dx_proportions_and_cnts.csv
   • codebooks/ccsr_codebook_for_sql.csv
   • codebooks/source materials/CCSR ICD10 categories 5.4.20_modified.csv
   • codebooks/base_score_modified.csv mapping of diagnosis entities in this study to the ocular emergencies surveyed in the BAsic SEverity Score for Common OculaR Emergencies (BaSe SCOrE) study by Bourges et al. (Supplementary Table 1 of the Supplementary Information)
   • codebooks/VT_vs_NVT_Big4_crosswalk.csv classification of diagnosis entities related to diabetic retinopathy, age-related macular degeneration, and glaucoma as vision-threatening vs. not vision-threatening (Supplementary Table 2 of the Supplementary Information)
     
     

   Outputs (supplementary data):

   • modeling-analysis/dx-entity-models-summstats.csv summary statistics and counterfactual model performance metrics for all 336 diagnosis entities considered for inclusion in the analysis Supplementary Data 2
   • modeling-analysis/dx-entity-codebook-export folder containing Word document tables (one for each diagnosis category) of the mapping between ICD-10-CM codes and diagnosis entities, derived from Supplementary Data 1
   • modeling-analysis/dx-entity-deviations.csv estimated deviations from expected utilization levels for each month of the pandemic study period, for the 261 diagnosis entities included in the analysis (visualized in Supplementary Figure 7 of the Supplementary Information)
   • modeling-analysis/dx-entity-deviation-bounds.csv 95% confidence intervals for estimated deviations from expected utilization levels for each month of the pandemic study period, for the 261 diagnosis entities included in the analysis (Supplementary Data 3)
   • modeling-analysis/dx-entity-deviation-pvalues.csv Unadjusted p-values for estimated deviations from expected utilization levels for each month of the pandemic study period, for the 261 diagnosis entities included in the analysis (Supplementary Data 3)
   • modeling-analysis/dx-entity-deviation-pvalues-adj.csv Adjusted p-values (via FDR) for estimated deviations from expected utilization levels for each month of the pandemic study period, for the 261 diagnosis entities included in the analysis (Supplementary Data 3)
   • modeling-analysis/dx-entity-quarterly-devs-and-recovery.csv summary statistics, 95% confidence intervals, p-values (unadjusted and adjusted) for estimated deviations from expected utilization levels for each quarter of the pandemic study period, as well as information on time-to-recovery and recovery status, for the 261 diagnosis entities included in the analysis (see the Fig 6 tab of Supplementary Data 4)
     
     

   Outputs (main figures):

   • main-figures/figure-2.pdf (Figure 2 of the Paper)
   • main-figures/figure-3.pdf (Figure 3 of the Paper)
   • main-figures/figure-4.pdf (Figure 4 of the Paper)
   • main-figures/figure-5.pdf (Figure 5 of the Paper)
   • main-figures/figure-6.pdf (Figure 6 of the Paper)
     
     

   Outputs (supplemental figures/tables):

   • suppl-figs-tables/figure-s1.pdf (Supplementary Figure 1 of the Supplementary Information)
   • suppl-figs-tables/figure-s2.pdf (Supplementary Figure 2 of the Supplementary Information)
   • suppl-figs-tables/figure-s3.pdf (Supplementary Figure 3 of the Supplementary Information)
   • suppl-figs-tables/figure-s4.pdf (Supplementary Figure 4 of the Supplementary Information)
   • suppl-figs-tables/figure-s5.pdf (Supplementary Figure 5 of the Supplementary Information)
   • suppl-figs-tables/figure-s6.pdf (Supplementary Figure 6 of the Supplementary Information)
   • suppl-figs-tables/figure-s7.pdf (Supplementary Figure 7 of the Supplementary Information)
   • suppl-figs-tables/table-s5.pdf (Supplementary Table 3 of the Supplementary Information)
   • suppl-figs-tables/table-s6.pdf (Supplementary Table 4 of the Supplementary Information)