Useful resources:

• Data Sharing Checklist_v2.0.pdf

• Cell size suppression examples and VRDC export guidance (external link)

What is cell size suppression?

A cell is the numeric value of a count, percentage, or range.

A cell size suppression policy prohibits sharing within a range or below a threshold with unauthorized individuals.  This is to protect patient, healthcare provider, healthcare facility, and/or state privacy.

Cell suppression or masking is a change that makes it impossible to determine a cell size below a permissible threshold.

• Counts of zero are permitted for all DataShare assets.

• Sometimes, multiple cells must be suppressed if together they can be used to determine the size of a small cell (i.e., back-calculation).  This applies to cells within the same table/figure and across different tables/figures/manuscripts.

Non-compliance with cell size suppression requirements may result in individual and institutional consequences, including fines and revocation of data access.

Cell Size Suppression Thresholds

Cell size suppression policies vary by data asset and are specified in the the data use agreement (DUA) between the data provider and the requesting institution (usually Duke).  Some data providers also post their policies online.

Cell size suppression requirements by data asset and who/what the cell represents (or from which it is derived):

* Unless linked to data from 1 - 10 insurance beneficiaries

** Obtain Duke Health Compliance and Duke Leadership before publishing provider or facility names.

Notes:

• Cells that contain results that must be suppressed are referred to as "small cells".

• Duke EHR’s cell suppression guidance is actively being revised; always check for the most recent guidance.

• Data providers that prohibit cell sizes of 1 – 10 often refer to this as “<11”.  However, zero is permitted for all data assets listed here.

• Get With The Guidelines (GWTG) data follows the CMS cell size suppression policy.

Practical Application

Small cells may manifest as counts, percentages, ranges, extreme values, and back-calculated values. 

Data assets that have suppression policies for patient, beneficiaries, and/or providers require masking for small cells of both living individuals and those who died.

CMS allows the reporting of p-values and the results of statistical models using small cells.

For PACE users: Raw and non-suppressed aggregate data is only permitted on the Linux server. Files moved to, or stored in Isilon folders (PACE project folder) must be aggregate data without small cells (i.e. suppression applied). 

Back-Calculation

Back-calculation allows a suppressed cell to be unmasked.  It occurs when two or more cells within a single table/figure or in different tables/figures together allow the viewer to calculate a masked value.

If a total N is reported, mask cells where N – value = small cell.

For categorical variables, if one category has a small cell, at least one other category must be masked.

Approaches to masking:

• Suppress an additional category

• Combine the category containing the small cell with another group so their total N is sufficiently large

• Coarsen (blur) data, i.e., report that the N for another group is larger than a value or within a range of values

Examples of these approaches are at https://resdac.org/articles/cms-cell-size-suppression-policy.

Extreme Values and Person-level Data

Cell suppression rules also apply to extreme values and small cells in figures.  Examples include:

• Outliers

• Maximums

• Minimums

• Censoring marks

• Risk set values

Linkage

Parallel data assets are analyzed independently.

Linked data assets generate results that are derived from both data assets.  The most stringent rule applies when linking data assets with differing cell suppression policies.

Linkage can also occur between types of source data within an asset, e.g., linking patients and providers from the same data asset.  If the rules for the data type differ within an asset, the most stringent rule applies.

Sneaky Places Small Cells Can Hide

• Consort tables (i.e., cohort inclusion/exclusion flowcharts)

• Survival curves

• Box-and-whisker plots

• Scatterplots

Best Practices

• Add a disclaimer to the statistical analysis plan’s cover page that results with small cells will not be reported or will be modified to comply with that data asset’s cell suppression policy.

• Use a data sharing checklist to make sure you are appropriately suppressing cells.

• Have the first author and statistician review the manuscript continually, especially the proof, for small cells using the checklist.

• Have someone double-check your work before sharing to those without permission to view small cells.

• Know the rules for your data.

• Know who has permission to view small cells.

• Footnote when suppressing cells.

• Indicate in a footnote whether the cell suppression policy applies when results from parallel data assets include a small cell and are shown beside results that require masking.

Sample Footnote Language

When result is from a data asset that requires suppression or is linked to an asset that requires suppression:

• Example 1: “This cell has been suppressed to protect patient privacy per the [data asset] cell suppression policy.”

• Example 2: "In accordance with the privacy policy of the [data provider], data from cells containing 1 -10 observations and data from cells that would allow for calculation of cells containing 1 -10 observations are not reported."

  • Replace with "1 - 9" for Duke EHR.

When showing results from a data asset without a cell size suppression requirement in the same table/figure with results from a data asset that does require suppression, it is helpful to clarify that the small cells are allowed.

Sample footnote: “Result derived exclusively from [data asset], which does not require cell suppression.”