Data and model bias in artificial intelligence for healthcare applications in New Zealand

Introduction

Developments in Artificial Intelligence (AI) are adopted widely in healthcare. However, the introduction and use of AI may come with biases and disparities, resulting in concerns about healthcare access and outcomes for underrepresented indigenous populations. In New Zealand, Māori experience significant inequities in health compared to the non-Indigenous population. This research explores equity concepts and fairness measures concerning AI for healthcare in New Zealand.

Methods

This research considers data and model bias in NZ-based electronic health records (EHRs). Two very distinct NZ datasets are used in this research, one obtained from one hospital and another from multiple GP practices, where clinicians obtain both datasets. To ensure research equality and fair inclusion of Māori, we combine expertise in Artificial Intelligence (AI), New Zealand clinical context, and te ao Māori. The mitigation of inequity needs to be addressed in data collection, model development, and model deployment. In this paper, we analyze data and algorithmic bias concerning data collection and model development, training and testing using health data collected by experts. We use fairness measures such as disparate impact scores, equal opportunities and equalized odds to analyze tabular data. Furthermore, token frequencies, statistical significance testing and fairness measures for word embeddings, such as WEAT and WEFE frameworks, are used to analyze bias in free-form medical text. The AI model predictions are also explained using SHAP and LIME.

Results

This research analyzed fairness metrics for NZ EHRs while considering data and algorithmic bias. We show evidence of bias due to the changes made in algorithmic design. Furthermore, we observe unintentional bias due to the underlying pre-trained models used to represent text data. This research addresses some vital issues while opening up the need and opportunity for future research.

Discussions

This research takes early steps toward developing a model of socially responsible and fair AI for New Zealand's population. We provided an overview of reproducible concepts that can be adopted toward any NZ population data. Furthermore, we discuss the gaps and future research avenues that will enable more focused development of fairness measures suitable for the New Zealand population's needs and social structure. One of the primary focuses of this research was ensuring fair inclusions. As such, we combine expertise in AI, clinical knowledge, and the representation of indigenous populations. This inclusion of experts will be vital moving forward, proving a stepping stone toward the integration of AI for better outcomes in healthcare.

A benchmarking study of home enteral nutrition services

BACKGROUND & AIMS

Patients receiving home enteral nutrition (HEN) via an enteral feeding tube often have complex healthcare requirements. There is limited information regarding how HEN care is provided within Australia and New Zealand. This study aimed to investigate the characteristics of HEN services and the provision of nutrition care to individuals receiving HEN within Australia and New Zealand.

METHODS

A cross-sectional study, surveying lead HEN dietitians for HEN services was conducted from the period 09 July 2019 to 20 September 2019 inclusive. An online survey was used to obtain data relating to the demographics, funding and clinical resources of respondents' HEN services. Services were benchmarked against a HEN service implementation checklist adapted from the Agency for Clinical Innovation (ACI).

RESULTS

Responses were received from 107 HEN services, with an estimated combined population of 7122 HEN patients. Services were predominantly government-funded (n = 102, 95.3%) and operated from acute hospitals (n = 57, 53.3%). The reported combined cost of all HEN equipment to the patient ranged from $0-$77 per week or $0-$341 per month. Fifty-two services were reported to have a dedicated HEN dietitian/coordinator, which was positively associated with the undertaking of quality improvement activities (p = 0.019). Mean compliance to the ACI HEN implementation checklist was 70.4% (±15.7%) with a range of 13.0-98.2%. Mean compliance was significantly higher in services with a HEN dietitian/coordinator than services without one (75.5% (±12.0%) vs 64.3% (±16.6%); p < 0.001).

CONCLUSIONS

This study provides detailed information regarding the characteristics of HEN services and nutrition care provided to enterally-fed patients across Australia and New Zealand. The majority of HEN services are not adhering to the ACI HEN service guidelines and there is considerable variation in cost burden for consumers indicating inequitable delivery of care to patients.

Nutrition management for critically and acutely unwell hospitalised patients with coronavirus disease 2019 (COVID-19) in Australia and New Zealand

Coronavirus disease 2019 (COVID‐19) results from severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2). The clinical features and subsequent medical treatment, combined with the impact of a global pandemic, require specific nutritional therapy in hospitalised adults. This document aims to provide Australian and New Zealand clinicians with guidance on managing critically and acutely unwell adult patients hospitalised with COVID‐19. These recommendations were developed using expert consensus, incorporating the documented clinical signs and metabolic processes associated with COVID‐19, the literature from other respiratory illnesses, in particular acute respiratory distress syndrome, and published guidelines for medical management of COVID‐19 and general nutrition and intensive care. Patients hospitalised with COVID‐19 are likely to have preexisting comorbidities, and the ensuing inflammatory response may result in increased metabolic demands, protein catabolism, and poor glycaemic control. Common medical interventions, including deep sedation, early mechanical ventilation, fluid restriction, and management in the prone position, may exacerbate gastrointestinal dysfunction and affect nutritional intake. Nutrition care should be tailored to pandemic capacity, with early gastric feeding commenced using an algorithm to provide nutrition for the first 5–7 days in lower‐nutritional‐risk patients and individualised care for high‐nutritional‐risk patients where capacity allows. Indirect calorimetry should be avoided owing to potential aerosol exposure and therefore infection risk to healthcare providers. Use of a volume‐controlled, higher‐protein enteral formula and gastric residual volume monitoring should be initiated. Careful monitoring, particularly after intensive care unit stay, is required to ensure appropriate nutrition delivery to prevent muscle deconditioning and aid recovery. The infectious nature of SARS‐CoV‐2 and the expected high volume of patient admissions will require contingency planning to optimise staffing resources including upskilling, ensure adequate nutrition supplies, facilitate remote consultations, and optimise food service management. These guidelines provide recommendations on how to manage the aforementioned aspects when providing nutrition support to patients during the SARS‐CoV‐2 pandemic.

A framework to support quality of care for patients with chronic intestinal failure requiring home parenteral nutrition

BACKGROUND AND AIM

Chronic intestinal failure requiring home parenteral nutrition (HPN) is a disabling condition that is best facilitated by a multidisciplinary approach to care. Variation in care has been identified as a key barrier to achieving quality of care for patients on HPN and requires appropriate strategies to help standardize management.

METHOD

The Australasian Society for Parenteral and Enteral Nutrition (AuSPEN) assembled a multidisciplinary working group of 15 clinicians to develop a quality framework to assist with the standardization of HPN care in Australia. Obstacles to quality care specific to Australia were identified by consensus. Drafts of the framework documents were based on the available literature and refined by two Delphi rounds with the clinician work group, followed by a further two involving HPN consumers. The Oxford Centre for Evidence-Based Medicine Levels of Evidence was used to assess the strength of evidence underpinning each concept within the framework documents.

RESULTS

Quality indicators, standards of care, and position statements have been developed to progress the delivery of quality care to HPN patients.

CONCLUSION

The quality framework proposed by AuSPEN is intended to provide a practical structure for clinical and organizational aspects of HPN service delivery to reduce variation in care and improve quality of care and represents the initial step towards development of a national model of care for HPN patients in Australia. While developed for implementation in Australia, the evidence-based framework also has relevance to the international HPN community.

Response and outcome from fluid resuscitation in acute pancreatitis: a prospective cohort study

BACKGROUND

Intravenous (IV) fluid resuscitation remains the cornerstone for early management of acute pancreatitis (AP), but many questions remain unanswered, including how to determine whether patients will benefit from additional fluids. The aim was to investigate the utility of serum biomarkers of responsiveness IV fluid resuscitation in patients with AP and systemic inflammatory response syndrome (SIRS).

METHODS

Eligible adult patients had abdominal pain for <36 h and ≥2 SIRS criteria. Mean arterial pressure (>65 mmHg) and urine output (>0.5 ml/kg/h) were used to assess responsiveness at 2 and 6-8 h after initiation of IV fluids. Comparison was made between responsive and refractory patients at time points for fluid volume, biomarkers and outcomes.

RESULTS

At 2 h 19 patients responded to fluids (Group 1) while 4 were refractory (Group 2); at 6-8 h 14 responded (Group 3) and 9 were refractory (Group 4). No demographic differences between patient groups, but Group 4 had worse prognostic features than Group 3. Refractory patients received significantly more fluid (Group 4 mean 7082 ml vs. Group 3 5022 mL, P < 0.001) in first 24 h and had worse outcome. No significant differences in biomarkers between the groups.

CONCLUSIONS

The serum biomarkers did not discriminate between fluid responsive and refractory patients. Refractory patients at 6-8 h had more severe disease on admission, did not benefit from additional fluids and had a worse outcome. New approaches to guide fluid resuscitation in patients with AP are required.

The oral refeeding trilemma of acute pancreatitis: what, when and who?

Tolerance of oral refeeding is an essential goal of nutritional management of acute pancreatitis. However, oral feeding intolerance remains one of the most common complications in patients with this disease. It often results in longer periods of hospitalization, increased treatment costs, increased risk of readmission, and reduced quality of life. The traditional practice involves keeping patients nil by mouth followed by gradual stepwise reintroduction of food. However, it does not have a solid evidence base and, hence, there is increasing interest in determining alternative strategies that may be beneficial in reducing the occurrence of oral feeding intolerance. This review focuses on the randomized controlled trials that investigated the key questions informing the nutritional management of acute pancreatitis: when to feed, what to feed and who is in charge of the decision-making.

Glutamine supplementation in acute pancreatitis: a meta-analysis of randomized controlled trials

BACKGROUND

There is emerging evidence that glutamine supplementation should be considered in patients with acute and critical illness associated with a catabolic response. There are reports of glutamine supplementation in acute pancreatitis but the results of these studies are conflicting. The aim of this study was to systematically review the randomised controlled trials (RCT) of glutamine in patients with acute pancreatitis.

METHODS

The Cochrane Central Register of Controlled Trials, MEDLINE, EMBASE, SCOPUS and 3 major Chinese databases were searched. The outcomes studied were mortality, total infectious complications, and length of hospital stay. A random effects model was used for meta-analysis of the outcomes in the included trials. A number of pre-specified subgroup analyses were also conducted. The summary estimates were reported as risk ratio (RR) for categorical variables and mean difference (MD) for continuous variables together with the corresponding 95% confidence interval.

RESULTS

Twelve RCT that enrolled 505 patients with acute pancreatitis were included in the final analysis. Overall, glutamine supplementation resulted in a significantly reduced risk of mortality (RR 0.30; 95% CI, 0.15 to 0.60; P < 0.001) and total infectious complications (RR 0.58; 95% CI, 0.39 to 0.87; P = 0.009) but not length of hospital stay (MD -1.35; 95% CI, -3.25 to 0.56, P = 0.17). In the subgroup analyses, only patients who received parenteral nutrition and those who received glutamine in combination with other immunonutrients demonstrated a statistically significant benefit in terms of all the studied outcomes.

CONCLUSIONS

This meta-analysis demonstrates a clear advantage for glutamine supplementation in patients with acute pancreatitis who receive total parenteral nutrition. Patients with acute pancreatitis who receive enteral nutrition do not require glutamine supplementation. Further studies are warranted to determine whether patients who receive combined enteral and parenteral nutrition need glutamine supplementation.