PTH-058 smart colonoscopy: using big-data to identify predictors of normal colonoscopic examinations

Stammers, Matt, Thalasekaran, Sreedhari and Bhandari, Pradeep (2018) PTH-058 smart colonoscopy: using big-data to identify predictors of normal colonoscopic examinations. Gut, 67 (1), A40-A41. (doi:10.1136/gutjnl-2018-BSGAbstracts.79).

Abstract

Introduction: endoscopy workload is increasing at a faster pace than available resources. The NHS has a wealth of data, which if used properly can improve resource allocation in future. The aim of this study was to review mass colonoscopy data to identify those factors most associated with a normal examination, in order to help rationalise future resource utilisation.

Methods: we constructed a standardised, anonymised database, containing all colonoscopies performed locally between 01/01/2010 and 10/12/2016. The records were then histology matched. The data was then analysed using the AnacondasTM 3 distribution of Python, using numpy, pandas, matplotlib and seaborn to clean and prepare, plot and perform statistical analysis on the data.

Results: 23 837 colonoscopies were performed on 18 489 individual adults during the study period. 544 procedures had to be excluded as they lacked an NHS number and couldn’t be histology matched. 23 293 procedures remained.
50.4% of the procedures were performed on females. The median age was 64. Across all the procedures, 25.46% were reported as entirely normal by the endoscopist. 3.04% of procedures contained a histologically confirmed cancer.
[PTH-058 Figure 1 Age vs% normal examinations not included].

Age: we found that the chances of obtaining a normal examination declined from ~49%±5% ≤43 years to ≤20%±2% in those ≥61 years. In patients aged ≤43 OR of a normal exam=3.29. For those aged ≥61, (OR=0.32 for a normal exam). Note, all OR’s in this study had p<0.0001 significance.

Sex: examinations performed on females were more likely to be reported as normal compared to men (OR=1.73). For women≤43, OR for normality=3.88.

Priority: routine priority was strongly associated with normal colonoscopy, (OR=1.99). Routine procedures on females≤43 were very likely to be normal (OR=4.90). These patients were very unlikely to have cancer (OR=0.093).

Indications: abdominal pain, anaemia (iron deficient) and bowel habit changes (of all types) and family history of colonic cancer were all found to be associated with a≥40% rate of normal examinations, (OR=3.57). The highest incidence of normal examinations was found amongst women≤43 undergoing routine colonoscopy for abdominal pain (OR=7.85), followed by bowel habit changes (OR=6.49), and anaemia (OR=5.91). Conversely, the highest rates of pathology were found in men≥61 undergoing bowel cancer screening, (OR for pathology=4.98; OR for malignancy=2).

Conclusions: we have developed a method for performing mass data analysis to identify trends in endoscopy data. Our data can help improve future utilisation of other colonic investigational modalities like colon capsule for low risk individuals. This can release colonoscopy capacity for the patients most at need.

This record has no associated files available for download.

Contributors

Author: Matt Stammers

Download statistics