Probabilistic topic models for unsupervised case identification Niklas NorĂŠn
Disclosures • Employed by the Uppsala Monitoring Centre • No specific funding received for this research
The case series
The case
The case
14 years Male Risperidone
1 mg / day 2014-11-02
Nausea Dizziness Vision abnormal
2014-12
How to find the case series
When we know what we are looking for‌
When we do not know what we are looking for‌
Different different but same
Nausea Headache Vision abnormal
Dizziness Insomnia Migraine Photopsia
Same same but different
Nausea Headache Vision abnormal
Nausea Vomiting Abdominal pain Constipation
Probabilistic topic model
0.45
0.35 0.20
Topic model Marginal probability for each ’topic’ 0.45
0.35 0.20
Topic model Marginal probability for each ‘topic’ 0.45
0.35 0.20
Probabilities for each medical event, per ‘topic’
Supervised learning
Supervised learning
Unsupervised learning
Unsupervised learning
Unsupervised learning
Unsupervised learning
Unsupervised learning
Unsupervised learning
Making sense of it all
Risperidone in VigiBase 1883 reports ~3.2 AEs Agitation
21%
Aggressive reaction
18%
Condition aggravated
16%
Psychosis
16%
…
…
1407 reports
1799 reports
~2.5 AEs
~2.8 AEs Extrapyramidal disorder
33%
Hyperprolacti- 76% naemia
Dystonia
20%
46%
Hyperkinesia
18%
Lactation nonpuerperal
Hypertonia
18%
Amenorrhoea
38%
…
…
…
…
Rivaroxaban in THIN 1199 patients ~2.7 Medical Events Phlebitis and thrombophlebitis
74%
Deep vein phlebitis 74% and thrombophlebitis of the leg Suspected deep vein 21% thrombosis
1684 patients ~2.3 Medical Events
529 patients
Cardiac dysrhythmias
100%
~2.5 Medical Events
Atrial fibrillation and flutter
99%
…
…
Acute pulmonary heart disease
97%
Pulmonary embolism 97% Diagnostic imaging
6%
Phlebitis and thrombophlebitis
6%
…
…
Oedema
18%
Leg swelling
15%
…
…
References 1.
Hand DJ, Bolton R. Pattern discovery and detection: a unified statistical methodology. Journal of Applied Statistics, 2004. 31(8):885-924.
2.
Norén GN, Fransson J, Juhlin K, Chandler R, Edwards IR. Adverse event cluster analysis for syndromic surveillance. Drug Safety 2015. 38:958 (Abstract).
3.
Orre R, Bate A, Norén GN, Swahn E, Arnborg S, Edwards IR. A Bayesian recurrent neural network for unsupervised pattern recognition in large incomplete data sets. International Journal of Neural Systems, 2005. 15(3):207-222.
4.
Chandler RE, Juhlin K, Fransson J, Caster O, Edwards iR, Norén GN. Current Safety Concerns with HPV vaccine: A Cluster Analysis of Reports in VigiBase. Drug Safety 2017; 40(1):81-90.