# Early-life risk factors predicting growth retardation and mortality in pigs: a multi-criteria approach

## TL;DR

This study identifies early-life risk factors in pigs that predict poor growth and mortality, using a predictive model based on birth-related traits to improve swine production efficiency.

## Contribution

A novel multi-criteria predictive model using early-life traits to classify pigs at risk of poor growth or mortality is developed and validated.

## Key findings

- Body weight on day 7 was the strongest predictor of weaning weight, with an R² > 0.60.
- The final model achieved an AUC of 0.910, using birth weight, relative birth weight, colostrum intake, and sow parity to predict compromised pigs.
- Pigs in the 10th percentile for birth weight with low colostrum intake had the highest risk of poor early-life performance.

## Abstract

Body weight (BW) variability throughout the production cycle remains a major challenge for the swine industry, particularly due to the negative impact of slow-growing pigs on production efficiency and batch uniformity. This study aimed to identify early-life risk factors associated with poor postnatal growth and mortality and to develop a multi-criteria predictive model for classifying pigs based on their early growth and survival potential. Data from 2,138 pigs (Pietrain × [Landrace × Yorkshire]) and 1,115 pigs (Pietrain × [Landrace × Large White]), collected from two commercial farms, were analyzed. Pigs were monitored from birth to weaning, with detailed records of farrowing traits, BW, body conformation indicators, colostrum intake (CI), and survival outcomes. Body weight on d 7 was the strongest predictor of weaning weight (R2 > 0.60), highlighting the critical influence of the first week on later growth. Logistic regression models were used to classify pigs as either compromised (defined as dead or alive with a BW below the 15th percentile of the BW distribution on d 7 of life and at weaning) or normal. The classification performance of the competing models, as well as the selection of the final model, was evaluated using the area under the curve (AUC) of the receiver operating characteristic curve. Subsequently, the optimal classification threshold was adjusted to balance sensitivity and positive predictive value. The final model, trained on d 7 data, achieved a high AUC (0.910), with BW on d 1, relative BW (RBW) on d 1, CI, and sow parity all significantly associated with the probability of being classified as compromised (P < 0.05). Each 100 g increase in BW on d 1 was associated with a 27.6% decrease in the odds of being compromised. Similarly, greater RBW on d 1 and CI were linked to a reduced risk. Pigs falling within the 10th percentile for BW on d 1, with low CI and negative RBW on d 1, showed the highest probability of being compromised by d 7. The model’s robustness was confirmed through consistent performance across datasets. Density plots further validated the model, illustrating clear distributional differences between compromised and normal pigs. These findings suggest that a model based on easily measurable birth-related indicators can reliably identify pigs at risk of poor early-life performance. Such a tool holds strong potential for on-farm application to enhance pig management and reduce BW variability at slaughter.

The development of a multi-criteria model based on easily measurable birth-related risk factors, with broad applicability across different pig populations, could enable the swine industry to identify pigs at risk of growth retardation or early mortality. In turn, this could facilitate timely interventions, improve efficiency, and reduce body weight variability at slaughter.