SARS-CoV-2 negative inpatient identification issue

Internal data quality checks have highlighted an issue with the identification of inpatient versus non-inpatient status among SARS-CoV-2 negative test results at some laboratories. This issue affects the PCR-based tests that we have linked to UK Biobank from Public Health England (field 40100).

For some laboratories, we report SARS-CoV-2 negative inpatients as non-inpatients (value 0 in the origin column).

The issue became apparent because we report large numbers of negative test results for some laboratories, none (or a very small number) of which are identified as inpatients. Notable laboratories include

• Northern General Hospital (Sheffield)
• St George’s Hospital Tooting
• Leeds General Infirmary
• Oxford John Radcliffe
• Darent Valley Hospital Dartford
• Royal Liverpool University Hospital

Additionally, we appear to have intermittently reported COVID-19 negative inpatients as non-inpatients for Sunderland Royal Infirmary. The identity of the laboratory is coded in the laboratory column. UK Biobank codes these seven laboratories as 38, 11, 31, 40, 19, 56 and 63 respectively.

We have tracked down the source of the issue to heterogeneity in the processing of negative results between laboratories. All laboratories report positive results to the Second Generation Surveillance System (SGSS). Usually only positive microbiological results are reported to SGSS. Exceptionally for SARS-CoV-2, negative results are also reported. This happens in two ways:

1. Directly to SGSS, with a pseudo-code indicating the organism (SARS-CoV-2) and test result (negative), where usually an organism alone is reported. Most laboratories have taken this approach.
2. Directly to Respiratory DataMart, a system set up to monitor influenza. The affected laboratories have taken this approach. We access the negative results for these laboratories separately, and do not import data required to identify inpatient status.

We are investigating a remedy for the issue. Meanwhile we advise researchers to be aware of the potential for artefacts, for example when comparing positive inpatients to negative inpatients.

(Strictly speaking, the coding in UK Biobank is correct:

• 0: No explicit evidence in microbiological record that the participant was an inpatient
• 1: Evidence from microbiological record that the participant was an inpatient

So the issue is one of heterogeneity of evidence, which the careful wording allows for.)