What is the value of statistical testing of observational data?

Machine-learning-based prediction of functional recovery in deep-pain-negative dogs after decompressive thoracolumbar hemilaminectomy for acute intervertebral disc extrusion

OBJECTIVE

To develop and compare machine-learning algorithms to predict recovery of ambulation after decompressive surgery for acute intervertebral disc extrusion (IVDE).

STUDY DESIGN

Multicenter retrospective cohort study.

SAMPLE POPULATION

Deep-pain-negative dogs with acute IVDE (n = 162).

METHODS

Clinical variables were preprocessed for machine learning and split into independent training and test sets in an 80:20 ratio. Each model was trained and internally validated on the full test set (Testfull) and the XGBoost algorithm validated on the same test set with preoperative variables withheld (Testwh).

RESULTS

Recovery of ambulation was recorded in 86/162 dogs (53.1%) in this sample population after decompressive surgery. The XGBoost algorithm achieved the best performance with an area under the receiver operating characteristic curve (AUC) of .9502 (95% CI: .8919-.9901), an accuracy of .8906 (95% CI: .8125-.9531), a sensitivity of .8750, and a specificity of .9063 on Testfull. XGBoost performance on Testwh was decreased, with an AUC of .8271 (95% CI: .7186-.9209), an accuracy of .7187 (95% CI: .6093-.8281), a sensitivity of .5625, and a specificity of .8750.

CONCLUSION

Machine-learning algorithms may predict outcomes accurately in deep-pain-negative dogs with IVDE after decompressive surgery. The XGBoost algorithm performed best on tabular data from this veterinary population undergoing spinal surgery.

CLINICAL SIGNIFICANCE

Machine-learning algorithms outperform current methods of prognostication. Pending external validation, machine-learning algorithms may be useful as assistive tools for surgical decision making.

Comparison of Survival After Treatment of Presumed Intracranial Meningioma by Radiotherapy or Surgery in 285 Dogs

BACKGROUND

The comparative effectiveness of radiotherapy and surgery for treating intracranial meningioma is unknown.

OBJECTIVES

To compare survival after treatment of suspected intracranial meningioma by either surgery or radiotherapy.

ANIMALS

Two hundred eighty-five companion dogs with suspected intracranial meningiomas presenting to 11 specialty clinics in three countries.

METHODS

Parallel cohort comparison study on retrospective data. Dogs diagnosed with intracranial meningioma by board-certified veterinary neurologists or radiologists and treated by radiotherapy or surgery were identified through medical record searches and presenting and survival data extracted. Lesion site was classified as rostro- or caudotentorial and size was measured on contrast magnetic resonance images. Outcome was all-cause death. Analysis of survival by Cox proportional hazards, including selection for optimal multivariable model using lasso, counterfactual modeling including variables associated with treatment allocation and survival.

RESULTS

One hundred sixty-eight dogs received radiotherapy and 117 received surgery. All analyses indicated reduced survival associated with surgery compared to radiotherapy. There was a median survival after surgery of 297 (IQR: 99-768) days compared with 696 (IQR: 368-999) for dogs treated by radiation, associated with a univariable hazard ratio of 1.802 (95% CI: 1.357-2.394). Counterfactual modeling estimated a mean survival of 480 (95% CI: 395-564) days after surgery and 673 (95% CI: 565-782) days after radiotherapy, representing a decrease in survival of 29%. Location and size of the lesion were not associated with survival duration.

CONCLUSIONS AND CLINICAL IMPORTANCE

Dogs with suspected intracranial meningioma have substantially superior survival after radiotherapy compared to surgery.

Influence of delayed milk ejection on mammary gland health and milking performance in dairy cows: A systematic review and meta-analysis

Delayed milk ejection (DME) could compromise udder health and milking performance. The objectives of this systematic review and meta-analysis were to (i) explore the current evidence, assess the risk of bias, and identify the knowledge gaps in published studies that investigated the effect of DME on udder health and milking performance in dairy cows, and (ii) quantify the magnitude of the difference in milk yield per milking session (MY), milking duration (MD), and peak milk flow rate (PFR) between cows with and without DME. We performed the current systematic review and meta-analysis according to PRISMA 2020 statement. We tracked the pertinent peer-reviewed publications from PubMed using the following syntax (delayed milk ejection OR disturbed milk ejection OR bimodal milk flow OR bimodality) AND (bovine OR cattle OR cows), and additional relevant studies from the identified articles' reference lists. We calculated the pooled-mean difference in MY (kg), MD (seconds), and PFR (kg/min) between cows with and without DME using the random-effects models. The selection process identified 15 studies as the final number of studies qualified for review, including six interventional and nine observational studies published between 1980 and 2023 conducted in seven countries, including the USA, Germany, Italy, Switzerland, Netherlands, New Zealand, and Portugal. The results indicated that while the DME had a minimal influence on the udder health, it can reduce MY by 1.55 kg per milking session (95 % CI = -2.10, -1.00). The observed evidence here suggests that DME can negatively influence the milking profitability.

Reporting and interpreting statistical results in veterinary medicine: Calling for change

Understanding and correctly interpreting statistical results presented in scientific articles is a required skill for practicing evidence-based veterinary medicine. A prerequisite for doing so is the adequate reporting of the results in scientific journals. However, most authors of veterinary publications determine the importance of their findings based on statistical significance (ie, P < .05), indicating that neither the limitations of using P values for inference nor the existence of more appropriate alternatives are widely appreciated in veterinary medicine. This deficiency in knowledge indicates a need to increase awareness in veterinary medicine regarding reporting statistical measures that quantify the magnitude of an effect along with its level of uncertainty, and then interpreting these results for clinical decision making. We utilize a hypothetical randomized controlled trial of dietary management in cats with chronic kidney disease to discuss some common misconceptions about P values and provide practical suggestions for alternatives. Reporting appropriate effect estimates along with their confidence intervals will allow veterinarians to easily and correctly determine whether the magnitude of the effect of interest meets clinical needs while acknowledging uncertainty in the results. We also describe confidence interval functions and show their utility as visual tools in aiding interpretation of confidence intervals. By providing practical guidance, we show that a change in reporting and interpreting statistical results is feasible and necessary. We hope this crucial step will promote clinical decision making based on effect estimates and confidence intervals.

Prevalence of portal vein and splanchnic venous thrombosis in dogs with chronic hepatitis

OBJECTIVES

Alterations in haemostasis have been described in dogs and humans with chronic hepatitis. Portal vein thrombosis is a recognised complication of chronic hepatitis in humans; however, its prevalence in dogs with chronic hepatitis has not been reported. We aimed to estimate the prevalence of, and describe clinical and laboratory data of dogs with chronic hepatitis and portal vein thrombosis and splanchnic venous thrombosis.

MATERIALS AND METHODS

Retrospective cross-sectional study. Medical records of dogs admitted to a veterinary teaching hospital between 2009 and 2019 were reviewed. Dogs were included if chronic hepatitis was histopathologically confirmed, and if diagnostic imaging or necropsy indicated the presence of thrombosis. Clinical and laboratory data (i.e. haematology, biochemistry, coagulation panels) were recorded. Descriptive statistics were used to characterise dogs with and without thrombosis.

RESULTS

Records from 136 dogs with chronic hepatitis were identified. Three of these dogs, 2.2% (95% confidence interval: 0.8 to 6.3%) all females, were diagnosed with portal vein thrombosis. Five dogs in total, (3.7%; 95% confidence interval: 1.6 to 8.3%), including three with portal vein thrombosis, all females, were diagnosed with splanchnic venous thrombosis. Dogs with portal vein and splanchnic venous thrombosis often had hyperbilirubinaemia, increased serum gamma-glutamyl transferase activity, and decreased plasma antithrombin 3 activity. They also had relatively high alternative Child-Pugh scores for dogs (median 6 out of 13).

CLINICAL SIGNIFICANCE

Portal vein and splanchnic venous thrombosis are potentially serious complications that were identified in a relatively low proportion of dogs with chronic hepatitis.