Dr Michelle Ananda-Rajah FRACP, PhD General Medicine & … · D +33 hepatosplenic lesions •...
Transcript of Dr Michelle Ananda-Rajah FRACP, PhD General Medicine & … · D +33 hepatosplenic lesions •...
Improving supportive care in cancer with innovative use of AI for invasive fungal diseases
Dr Michelle Ananda-Rajah FRACP, PhDGeneral Medicine & Infectious Diseases Alfred Health, [email protected]
@rajah_mich
Invasive mold diseases cause a life threatening pneumonia
57 female, Burmese, MDS, aspergillosis
69 male, Richter’s transformation (CLLDLBCL), Aspergillus flavus, fumigatus
1Lortholary et al CMI 2011, Kontoyiannis et al CID 2010
Antifungal stewardship: why does it matter?• Poor clinical outcomes & cost associated with IFD
– Mortality 40-90%1,2
– Attributable hospitalization cost is high AUD$79K (2011)3
• 1Kontoyiannis et al CID 2010, Even et al Haematologica 2011, 3Ananda-Rajah et al AAC 2011
• Overuse of AF drugs is an issue1
– Poor diagnostics– High cost drugs
• 1Ananda-Rajah et al Curr Opin Inf Dis 2011, 2Shah et al JAC 2011
• Toxicities (short/long term), drug-drug interactions are significant
• Local=institutional, epidemiology is variable but highly relevant– IFD incidence 3-21% (US)1; 0-29% (China)2
• 1Kontoyiannis et al CID 2010 2Sun et al Biol Bone Marrow Transplant 2015
Case 1: 23yr old female
• High grade Burkitts with CNS disease• D+11 chemo, febrile neutropenia, 39C• D+16, ICU admit, delirium, sepsis
• D18+ ? Antifungal prophylaxis
D+20 chemotherapy
Brain and sinuses CT:• mucosal thickening• Bony deficiency• Possibility of aggressive
fungal sinus infection
D+26 sinus biopsy
• Histo amendment– possibility of a mixed fungal
population (?mucormycosis, aspergillus, scedosporium)
• Alternaria spp
D +33 hepatosplenic lesions• Disseminated disease with
multiple intra-abdominal lesions
• (System) failure of antifungal stewardship
Antifungal stewardship: where are the gaps?
AFS performance measures: a systematic review • “ Antifungal consumption appears to be the most achievable
performance measure to evaluate the impact of an antifungal stewardship program”
• Reductions in antifungal drug consumption– 12-71%
• Reductions in antifungal cost– 9.7% to 50%
• Data on quality of care??–mortality, fungal incidence/spectrum, patient groups, effectiveness of prophylaxis
Bienvenu et al JAC 2018
Antifungal stewardship where to start?British Society for Antimicrobial Chemotherapy guidelines
Enoch et al, 2018
Surveillance & clinical audit:• Identify gaps• Areas that need
closer attention
Surveillance of fungal diseases is not easy…
Molds have no reliable lab marker: micro positive in <50% pts
• Blood cultures negative
• Culture & histopathology positive in ~50%• Hope et al Lancet ID 2005
• Biomarkers unreliable– Galactomannan Sensitivity 71% Specificity 86%
Pfeiffer et al CID, 2006
• PCR is investigational– PCR Sensitivity 75% Specificity 87%
• Mengoli et al Lancet ID, 2009
The diagnostic pathway leads through CTMolds present as pneumonia in 90-100%1
57 female, Burmese, MDS, aspergillosis
69 male, Richter’s transformation (CLLDLBCL), Aspergillus flavus, fumigatus
1Lortholary et al CMI 2011, Kontoyiannis et al CID 2010
CT reportCTCHEHICT Chest Hi Resolution XX/XX/XX at 1202REPORT:History.Neutropenic sepsis. Increasing temperature. Unwell.? Invasive fungalinfection.
Findings:There are poorly-defined scattered alveolar densities in the rightupper lobe with at least 3 nodular densities with the largest measuring1 cm. There is less marked involvement of the right lower and middlelobe as well as the left lung. There is a predeliction for theseabnormalities along the bronchovascular bundle.There is a moderate right and a small left effusion.There are no axillary or mediastinal lymph nodes.Conclusion:Widespread bilateral alveolar opacity associated with multiple noduleswith more marked involvement of the right upper lobe. These findingsare highly suggestive of invasive fungal infection.Invasive bronchopulmonary aspergillosis needs to be considered.Reported by: Dr XXXX
2014
Natural language processing for fungal infectionsCharacteristic TP FP TN FN Sn, % Sp, %
Training scans, n=366
197 32 117 20 91 79
Held-out scans, n=83
35 13 30 5 88 70
Ananda-Rajah et al Plos One 2014
Report levelSn 91%, Sp 79%
Patient level n=188/393 pts, 1247 reports100% sensitivity/recall
ALTA, 2017
A decade of Alfred data for entire haem population
June2008-Dec 2017
• 229 pts , 247 episodes
• On trial drugs =21%
• Outpt episodes 6%
Underlying disease %
AML 56
ALL 13
NHL 9.7
Myeloma 7.7
Myelodysplastic syndrome 3.2
Hodgkins Lymphoma 2.8
Chronic lymphocytic leuk/small lymphocytic lymphoma
2.8
Acute promyelocytic leuk N=2
CML N=2
HSCT, n=83 34%
Allogeneic 77%
Autologous 23%
Early post HSCT (<100d) 40%
Late HSCT (>100d) 60%Baggio et al Immunocompromised Host Society Symposium, Athens 2018
AFS: gaps, missed opportunities• Breakthrough infections n=139 episodes, 56%
-posaconazole TDM in 2 wks prior 60%, on target in 48%-voriconazole TDM in 2wks prior 53%, on target in 82%
• Antifungal prophylaxis indicated but not given in 2 wksprior– Early HSCT (<100d) in 4 episodes– HSCT with GVHD within 60 days of IMD in 16 episodes– Persistent neutropenia >5 wks in 9 episodes– Persistent neutropenia >3 wks but <5 wks in 10 episodes
• Emerging groups lacking evidence for prophylaxis, 25%– ALL, myeloma, CLL, APML, MDS, lymphoma, aplastic anaemia
Convolutional neural networks for image recognition of fungal pneumonia
Ananda-Rajah et al Immunocompromised Host Society Symposium, Athens 2018
Case 1: 72 yr old man AML on trial drug
Radiologist MACHINE
Radiologist MACHINE
Radiologist MACHINE
Radiologist MACHINE
Radiologist MACHINE
HUMAN MACHINE
Is more data better? The expert system
JCO Clin Cancer Informatics 2017
The “safety net” for cases potentially missed by natural language processing
Ananda-Rajah et al JCO CCI 2017
Ananda-Rajah et al JCO CCI 2017
Ananda-Rajah et al JCO CCI 2017
Expert system: Sensitivity maintained, with 48% reduction in false positives
Ananda-Rajah et al JCO CCI 2017
www.fungalai.com
Next steps…prospective multicenter study
• Alfred Health• Eastern Health• Monash Medical Centre• Monash Childrens• Westmead Hospital• Fiona Stanley Hospital• Royal Adelaide Hospital
Ensuring value… fungalAi as an enabler
• To enable…
• Prospective surveillance• Antifungal stewardship• Clinical risk stratification “big
data” • fungal reporting in clinical
trials• Business case proposals • Cost effectiveness analyses• Improved coding• Decision support for
radiologists
• To benefit…• Patients: optimize outcomes!
tailor prophylaxis, regional areas, new risk groups
• Clinicians: strengthen antifungal stewardship, outbreak detection
• Radiologists: decision support• Researchers: discovery• Trial sponsors: adverse events• Hospitals: reimbursement
• Prof Geoff Webb • Prof Tom Drummond • Dr Reza Haffari• Drs Titus Tang, Horace Josh• Drs Francois Pettitjean, Christoph
Bergemeir
Funders• The Alfred Foundation • Monash Partners• Monash Inst Medical Engineering MIME• Faculty Medicine Nursing Health
Sciences• Monash University
• A/Prof Dinesh Varma • Dr Samantha Ellis• Dr Anthony Kam• Dr Jarrel Seah• Dr Trisha Peel• Prof Anton Peleg • A/Prof Andrew Wei• Ms Ann Larkins, Prof Andrew Way
fungalAi.com
QUESTIONS/COMMENTS ?
@rajah_mich