Carbon monoxide poisoning surveillance in the Veterans Health Administration, 2010–2017

We analyzed VA electronic medical record (EMR) data for inpatient discharges and outpatient encounters from January 2010 through December 2017 with ICD-9-CM or ICD-10-CM codes meeting the 2018 Council of State and Territorial Epidemiologists (CSTE) criteria for CO poisoning case ascertainment using administrative data [19]. Veterans receiving care at non-VA healthcare facilities during the study period and whose coded encounters met CSTE criteria for CO poisoning were included. Patients with multiple visits classified as CO poisoning during the study period were counted once, for their initial encounter. Demographic, laboratory, and treatment data, and date of death (where applicable) recorded in the EMR for each unique patient with CO poisoning for the study time-period was collected. CO poisoning surveillance was conducted as part of VHA public health operations. As such, the Veterans Health Administration Office of Research Oversight considers public health investigations as operational activities and not research in VHA [20].

If patients had outpatient and inpatient codes for the same day, one unique encounter was counted, with highest level of care (i.e., inpatient) selected. To reduce inclusion of patients whose ICD-9-CM code was not indicative of initial exposure, we reviewed patient problem lists and removed 34 patients whose coded problem list indicated exposure to CO occurred prior to the study period. For patient encounters with ICD-10-CM codes, we reviewed the seventh character extension to determine if the encounter was deemed initial, subsequent, or related to CO poisoning sequela and a query was performed to determine if any of these patients had CO poisoning-coded visits prior to 2010. One patient with an ICD-10-CM sequela episode of care extension had an initial encounter in 2004 and was removed.

Based on 2018 CSTE criteria utilizing ICD-9-CM and ICD-10-CM codes, unique patients were classified by degree of certainty that poisoning was CO-related (confirmed, probable, suspected). A case was considered confirmed if the inpatient discharge or outpatient encounter was coded with ICD-9-CM code 986 or ICD-10-CM code T58 (toxic effect of carbon monoxide); or with an ICD-9-CM External Cause of Injury code (E-code) explicitly indicating CO exposure was present in the absence of code 986 (E868.3, E868.8, E868.9, E952.1, E982.1). Probable cases were defined as any encounter with ICD-9-CM E-code indicating motor vehicle exhaust exposure (E868.2, E952.0, E982.0). Cases were further categorized by intent as unintentional, assault, intentional self-harm, or undetermined based upon ICD-9-CM E-codes or specific ICD-10-CM codes pertaining to the nature of CO poisoning exposure. The full list of CSTE ICD-9-CM and ICD-10-CM codes and frequency of occurrence among cases are listed in Additional file 1; Table S1.

We utilized Logical Observation Identifiers Names and Codes (LOINC) listed by the Public Health Information Network Vocabulary Access and Distribution System [21] as laboratory tests associated with CO poisoning (32160–4, 31157–1, 20563–3, 2030–5, 2031–3, 41648–7, 2032–1, and 2029–7) to determine whether carboxyhemoglobin (COHb) blood testing of patients coded for CO poisoning by ICD-9-CM or ICD-10-CM was performed. A COHb measurement was considered collected at time of encounter if the date of COHb sample collection was within 0–2 days before or after encounter date. 2018 CSTE laboratory criteria for confirmed/probable classification was based on COHb blood level > 9% for probable or > 12% for confirmed cases in settings where smoking history is unknown [19]. We analyzed procedure codes ICD-9-CM 93.95 and the ICD-10 Procedure Coding System codes 5A05121 and 5A05221 for hyperbaric oxygen (HBO₂) therapy and dates of death within intervals of 30 days or 1 year relative to CO poisoning encounter dates as outcome data for patients with confirmed, probable, and suspected CO poisoning.

We analyzed demographic variables and compared to unique users of VHA care during the 2010–2017 time-period, calculating 95% CIs. Seasonality trends for CO poisoning encounters were analyzed across the study period. Rates were calculated using total unique users of VHA care for matching time frame and VHA-defined demographic category as denominator. Rates of CO poisoning by U.S. Census region [22] were determined based on state location of the parent facility where the encounter occurred, using number of unique users of VHA care per state. Rates for confirmed/probable cases were compared to those of suspected cases across all variables looking for potential differences based on case categorization and testing for statistical significance using the two-tailed z test for proportions. Statistical analysis was performed using Python Statsmodels module, version 0.9.0 [23].

We performed EMR review for 678 patients including a) all 595 confirmed/probable CO poisoning cases from 2010 to 2016 initially categorized with undetermined injury intent; b) a random sample of 20 patients out of 1847 (1%) with suspected CO poisoning coded with ICD-9-CM code E825.x for “Other motor vehicle non-traffic accident of other and unspecified nature”; c) a random sample of 20 confirmed/probable cases with source of CO poisoning “other or unspecified”, d) all 26 patients with confirmed/probable CO poisoning who died within 30 days of CO poisoning encounter; and e) all 17 patients who received HBO₂ therapy.

Our study has several limitations. In limiting cases to those coded with ICD-9-CM or ICD-10-CM CO poisoning codes, we may have omitted those miscoded or not coded. In addition, clinicians may not have considered CO poisoning when patients presented to the ED. Previous studied have found that 3–5% of patients presenting with headache and dizziness had occult CO poisoning [36], and therefore could have contributed to fewer cases found in our study. We did not include National Death Index data, so patients who died from their CO poisoning before receiving treatment at a VHA hospital are not included in our study. While we attempted to limit inclusion of miscoded cases caused by carry-over of historic exposure codes, we cannot be sure that all cases represented current diagnoses of CO poisoning. Conversely, our method of limiting cases to initial occurrence only may have eliminated cases that represented true second exposures, for example, repeat suicide attempts or unintentional exposures in homes where original source of CO has not been remediated. Our review of cases classified as “undetermined” indicates misuse of this category code may have led to underrepresentation of unintentional and intentional injury intent categories of CO poisoning in our study. Similarly, we found a high percentage of cases coded with an “other or unspecified” source of CO poisoning. Our evaluation of COHb blood values was likely limited by the fact that initial exposure to CO is often measured at the scene of CO exposure by first responders using pulse CO-oximetry. These values, and those of patients transferred to non-VA emergency care are scanned into the Veterans’ EMR and thus not available as discrete, electronically queriable LOINC codes. Our review of the 17 cases treated with HBO₂ indicates that COHb levels at time of treatment and recorded within scanned documents were much higher than measurements available in the EMR for those patients. Capture of HBO₂ treatment may be similarly limited as it is routinely performed at non-VA sites.