Impact of primary to secondary care data sharing on care quality in NHS England hospitals

Data Quality-Driven Improvement in Health Care: Systematic Literature Review

BACKGROUND

The promise of real-world evidence and the learning health care system primarily depends on access to high-quality data. Despite widespread awareness of the prevalence and potential impacts of poor data quality (DQ), best practices for its assessment and improvement are unknown.

OBJECTIVE

This review aims to investigate how existing research studies define, assess, and improve the quality of structured real-world health care data.

METHODS

A systematic literature search of studies in the English language was implemented in the Embase and PubMed databases to select studies that specifically aimed to measure and improve the quality of structured real-world data within any clinical setting. The time frame for the analysis was from January 1945 to June 2023. We standardized DQ concepts according to the Data Management Association (DAMA) DQ framework to enable comparison between studies. After screening and filtering by 2 independent authors, we identified 39 relevant articles reporting DQ improvement initiatives.

RESULTS

The studies were characterized by considerable heterogeneity in settings and approaches to DQ assessment and improvement. Affiliated institutions were from 18 different countries and 18 different health domains. DQ assessment methods were largely manual and targeted completeness and 1 other DQ dimension. Use of DQ frameworks was limited to the Weiskopf and Weng (3/6, 50%) or Kahn harmonized model (3/6, 50%). Use of standardized methodologies to design and implement quality improvement was lacking, but mainly included plan-do-study-act (PDSA) or define-measure-analyze-improve-control (DMAIC) cycles. Most studies reported DQ improvements using multiple interventions, which included either DQ reporting and personalized feedback (24/39, 61%), IT-related solutions (21/39, 54%), training (17/39, 44%), improvements in workflows (5/39, 13%), or data cleaning (3/39, 8%). Most studies reported improvements in DQ through a combination of these interventions. Statistical methods were used to determine significance of treatment effect (22/39, 56% times), but only 1 study implemented a randomized controlled study design. Variability in study designs, approaches to delivering interventions, and reporting DQ changes hindered a robust meta-analysis of treatment effects.

CONCLUSIONS

There is an urgent need for standardized guidelines in DQ improvement research to enable comparison and effective synthesis of lessons learned. Frameworks such as PDSA learning cycles and the DAMA DQ framework can facilitate this unmet need. In addition, DQ improvement studies can also benefit from prioritizing root cause analysis of DQ issues to ensure the most appropriate intervention is implemented, thereby ensuring long-term, sustainable improvement. Despite the rise in DQ improvement studies in the last decade, significant heterogeneity in methodologies and reporting remains a challenge. Adopting standardized frameworks for DQ assessment, analysis, and improvement can enhance the effectiveness, comparability, and generalizability of DQ improvement initiatives.

Health data sharing attitudes towards primary and secondary use of data: a systematic review

Background

To receive the best care, people share their health data (HD) with their health practitioners (known as sharing HD for primary purposes). However, during the past two decades, sharing for other (i.e., secondary) purposes has become of great importance in numerous fields, including public health, personalized medicine, research, and development. We aimed to conduct the first comprehensive overview of all studies that investigated people's HD sharing attitudes-along with associated barriers/motivators and significant influencing factors-for all data types and across both primary and secondary uses.

Methods

We searched PubMed, MEDLINE, PsycINFO, Web of Science, EMBASE, and CINAHL for relevant studies published in English between database inception and February 28, 2023, using a predefined set of keywords. Studies were included, regardless of their design, if they reported outcomes related to attitudes towards sharing HD. We extracted key data from the included studies, including the type of HD involved and findings related to: HD sharing attitudes (either in general or depending on type of data/user); barriers/motivators/benefits/concerns of the study participants; and sociodemographic and other variables that could impact HD sharing behaviour. The qualitative synthesis was conducted by dividing the studies according to the data type (resulting in five subgroups) as well as the purpose the data sharing was focused on (primary, secondary or both). The Newcastle-Ottawa Scale (NOS) was used to assess the quality of non-randomised studies. This work was registered with PROSPERO, CRD42023413822.

Findings

Of 2109 studies identified through our search, 116 were included in the qualitative synthesis, yielding a total of 228,501 participants and various types of HD represented: person-generated HD (n = 17 studies and 10,771 participants), personal HD in general (n = 69 studies and 117,054 participants), Biobank data (n = 7 studies and 27,073 participants), genomic data (n = 13 studies and 54,716 participants), and miscellaneous data (n = 10 studies and 18,887 participants). The majority of studies had a moderate level of quality (83 [71.6%] of 116 studies), but varying levels of quality were observed across the included studies. Overall, studies suggest that sharing intentions for primary purposes were observed to be high regardless of data type, and it was higher than sharing intentions for secondary purposes. Sharing for secondary purposes yielded variable findings, where both the highest and the lowest intention rates were observed in the case of studies that explored sharing biobank data (98% and 10%, respectively). Several influencing factors on sharing intentions were identified, such as the type of data recipient, data, consent. Further, concerns related to data sharing that were found to be mutual for all data types included privacy, security, and data access/control, while the perceived benefits included those related to improvements in healthcare. Findings regarding attitudes towards sharing varied significantly across sociodemographic factors and depended on data type and type of use. In most cases, these findings were derived from single studies and therefore warrant confirmations from additional studies.

Interpretation

Sharing health data is a complex issue that is influenced by various factors (the type of health data, the intended use, the data recipient, among others) and these insights could be used to overcome barriers, address people's concerns, and focus on spreading awareness about the data sharing process and benefits.

Funding

None.

Artificial Intelligence and Panendoscopy-Automatic Detection of Clinically Relevant Lesions in Multibrand Device-Assisted Enteroscopy

Device-assisted enteroscopy (DAE) is capable of evaluating the entire gastrointestinal tract, identifying multiple lesions. Nevertheless, DAE's diagnostic yield is suboptimal. Convolutional neural networks (CNN) are multi-layer architecture artificial intelligence models suitable for image analysis, but there is a lack of studies about their application in DAE. Our group aimed to develop a multidevice CNN for panendoscopic detection of clinically relevant lesions during DAE. In total, 338 exams performed in two specialized centers were retrospectively evaluated, with 152 single-balloon enteroscopies (Fujifilm®, Porto, Portugal), 172 double-balloon enteroscopies (Olympus®, Porto, Portugal) and 14 motorized spiral enteroscopies (Olympus®, Porto, Portugal); then, 40,655 images were divided in a training dataset (90% of the images, n = 36,599) and testing dataset (10% of the images, n = 4066) used to evaluate the model. The CNN's output was compared to an expert consensus classification. The model was evaluated by its sensitivity, specificity, positive (PPV) and negative predictive values (NPV), accuracy and area under the precision recall curve (AUC-PR). The CNN had an 88.9% sensitivity, 98.9% specificity, 95.8% PPV, 97.1% NPV, 96.8% accuracy and an AUC-PR of 0.97. Our group developed the first multidevice CNN for panendoscopic detection of clinically relevant lesions during DAE. The development of accurate deep learning models is of utmost importance for increasing the diagnostic yield of DAE-based panendoscopy.