Chief Scientific Officer CERobs Consulting Philadelphia, United States
Background: Metformin-induced lactic acidosis (LA) is a rare but serious adverse event (AE) among patients with diabetes. Claims-based studies of this AE are complicated by the fact that neither ICD-9-CM nor ICD-10-CM coding systems have a dedicated code for LA. Previous studies have relied on inpatient codes for acidosis (ICD-9-CM: 276.2; ICD-10-CM: E87.2) as a proxy, although only a fraction of these acidosis codes represent LA. These studies are further complicated by LA due to other causes (such as sepsis or malignancy), which is common during hospitalization.
Objectives: To construct a validated algorithm for hospitalization due to LA.
Methods: We identified US Medicare patients age ≥65 with diabetes who had a hospitalization with an inpatient ICD-9-CM or ICD-10-CM diagnosis code for acidosis in any position between 1/1/2007 and 12/31/2018. For our algorithm, we required both an admitting diagnosis of acidosis to ensure presence on admission and a discharge diagnoses of acidosis to assure a confirmed diagnosis during the hospital stay.
We requested all medical charts meeting these criteria and nephrologists reviewed the retrieved charts. Confirmed LA was defined by elevated lactate (≥3mmol/L) and evidence of acidosis (pH < 7.35). Probable LA was defined by elevated lactate without laboratory evidence of acidosis. Positive predictive values (PPV) and 95% confidence intervals [CI] were calculated separately for hospitalizations using ICD-9-CM and ICD-10-CM codes, as well as limited to principal discharge diagnoses (to denote clinical relevance) versus all discharge diagnoses.
Results: Of 145,389 patients with inpatient diagnosis of acidosis, 1,209 (0.8%) had an admitting and discharge diagnosis of acidosis. Of those, 183 (15%) were retrieved and adjudicated.
In 33 charts with admitting+principal ICD-10-CM diagnoses for acidosis, the PPV was 67% (CI 51%-83%) for confirmed LA and 82% (CI 69%-95%) for confirmed+probable LA. If discharge diagnoses in all positions were included (86 charts), the PPVs for confirmed and confirmed+probable decreased to 52% (CI 38%-66%) and 69% (CI 56%-82%), respectively.
In 55 charts with admitting+principal ICD-9-CM diagnoses for acidosis, the PPV was 38% (CI 25%-51%) for confirmed LA and 53% (CI 40%-66%) for confirmed+probable LA. If discharge diagnoses in all positions were included (97 charts), the PPVs for confirmed and confirmed+probable were 38% (CI 24%-52%) and 48% (CI 33%-62%), respectively.
Conclusions: In the absence of an LA-specific ICD code, our ICD-10-CM algorithm (admitting+principal discharge diagnoses for acidosis) can provide an adequate PPV for studies of metformin-induced LA in older adults in Medicare. However, the performance decreases substantially in ICD-9-CM or when including all discharge diagnoses.
