Emulating the EPIC trial using VetCompass primary-care data: causal effects of pimobendan in UK dogs with grade IV/VI heart murmurs

Abstract

Target trial emulation applies design principles from randomised controlled trials (RCTs) to the analysis of observational data, potentially replicating RCT results in real-world settings. The EPIC trial reported that pimobendan delays the onset of congestive heart failure (CHF) and extends survival in dogs with preclinical degenerative mitral valve disease (DMVD). The current study aimed to explore the extent to which target trial emulation approximates the EPIC trial results in a primary-care setting. Grade IV/VI murmur diagnosis was defined as the treatment intervention stage. There were 928 dogs ≥ 6 years and ≤ 15 kg at first grade IV/VI murmur diagnosis recorded from January 1, 2016, to December 31, 2018 in the VetCompass database included in the study. A causal inference "target trial emulation" approach using VetCompass anonymised clinical data was designed to replicate the EPIC trial with adaptation for a primary-care setting and to address immortal time bias, confounding bias and loss to follow-up. After bias adjustments to establish causal effects using observational data, the 5-year CHF cumulative incidence was lower in dogs prescribed pimobendan (34.1%, 95% CI 26.5-42.0) than dogs not prescribed pimobendan (56.3%, 95% CI 52.8-59.8). Dogs prescribed pimobendan had 311 fewer days of health lost to CHF (95% CI 224-395 days) within 5 years. Dogs prescribed pimobendan lived longer (adjusted mean survival time 1051 days, 95% CI 967-1125) than dogs not prescribed pimobendan (905 days, 95% CI 871-940 days). This study demonstrates that target trial emulation within veterinary research can replicate findings from RCTs. Clinically, the current findings suggest that preclinical grade IV murmur diagnosis may offer an appropriate intervention stage to begin pimobendan therapy in dogs with presumed DMVD.

Fig 1. Directed acyclic graph (DAG), based on existing published evidence and expert knowledge, used to identify factors to control for when evaluating the effect of pimobendan prescription for degenerative mitral valve disease in dogs on congestive heart failure.

The same causal structure was assumed to be sufficient to de-confound all-cause mortality as the outcome of interest.

Fig 2. Weighted cumulative incidence, over a maximum 5-year follow-up period, estimated using the Aalen–Johansen estimator for congestive heart failure (including cumulative incidence curves for death as a competing event) if dogs attending primary-care practices in the UK were prescribed pimobendan and if dogs were not prescribed pimobendan within 6 months of first grade IV heart murmur diagnosis.

Fig 3. Weighted Kaplan Meier estimation (over a maximum 5-year follow-up period) of survival curves for all-cause mortality in dogs attending primary-care practices in the UK if prescribed pimobendan and if dogs not prescribed pimobendan within 6 months of first grade IV/VI heart murmur diagnosis.

Fig 4. Standardised differences (%) for the outcome of congestive heart failure (with death a competing event) on the cloned data at six months after grade IV/VI murmur diagnosis, while accounting, or not, for selection bias (i.e., weighted and unweighted, respectively).

Fig 5. Standardised differences (%) for all-cause mortality as the outcome on the cloned data at six months after grade IV/VI murmur diagnosis, while accounting, or not, for selection bias (i.e., weighted and unweighted, respectively).