Prompts, Pearls, Imperfections: Comparing ChatGPT and a Human Researcher in Qualitative Data Analysis.

The impact of ChatGPT and other large language model-based applications on scientific work is being debated across contexts and disciplines. However, despite ChatGPT's inherent focus on language generation and processing, insights regarding its potential for supporting qualitative research and analysis remain limited. In this article, we advocate for an open discourse on chances and pitfalls of AI-supported qualitative analysis by exploring ChatGPT's performance when analyzing an interview transcript based on various prompts and comparing results to those derived by an experienced human researcher. Themes identified by the human researcher and ChatGPT across analytic prompts overlapped to a considerable degree, with ChatGPT leaning toward descriptive themes but also identifying more nuanced dynamics (e.g., 'trust and responsibility' and 'acceptance and resistance'). ChatGPT was able to propose a codebook and key quotes from the transcript which had considerable face validity but would require careful review. When prompted to embed findings into broader theoretical discourses, ChatGPT could convincingly argue how identified themes linked to the provided theories, even in cases of (seemingly) unfitting models. In general, despite challenges, ChatGPT performed better than we had expected, especially on identifying themes which generally overlapped with those of an experienced researcher, and when embedding these themes into specific theoretical debates. Based on our results, we discuss several ideas on how ChatGPT could contribute to but also challenge established best-practice approaches for rigorous and nuanced qualitative research and teaching.

Making the health system work for over 25 million births annually: drivers of the notable decline in maternal and newborn mortality in India.

INTRODUCTION: India's progress in reducing maternal and neonatal mortality since the 1990s was faster than the regional average. We systematically analysed how national health policies, services for maternal and newborn health, and socioeconomic contextual changes, drove these mortality reductions. METHODS: The study's mixed-methods design integrated quantitative trend analyses of mortality, intervention coverage and equity since the 1990s, using the sample registration system and national surveys, with interpretive understandings from policy documents and 13 key informant interviews. RESULTS: India's maternal mortality ratio (MMR) declined from 412 to 103 maternal deaths per 100 000 live births between 1997-1998 and 2017-2019. The neonatal mortality rate (NMR) declined from 46 to 22 per 1000 live births between 1997 and 2019. The average annual rate of mortality reduction increased over time. During this period, coverage of any antenatal care (57%-94%), quality antenatal care (37%-85%) and institutional delivery (34%-90%) increased, as did caesarean section rates among the poorest tertile (2%-9%); these coverage gains occurred primarily in the government (public) sector. The fastest rates for increasing coverage occurred during 2005-2012.The 2005-2012 National Rural Health Mission (which became the National Health Mission in 2012) catalysed bureaucratic innovations, additional resources, pro-poor commitments and accountability. These efforts occurred alongside smaller family sizes and improvements in macroeconomic growth, mobile and road networks, women's empowerment, and nutrition. These together reduced high-risk births and improved healthcare access, particularly among the poor. CONCLUSION: Rapid reduction in NMR and MMR in India was accompanied by increased coverage of maternal and newborn health interventions. Government programmes strengthened public sector services, thereby expanding the reach of these interventions. Simultaneously, socioeconomic and demographic shifts led to fewer high-risk births. The study's integrated methodology is relevant for generating comprehensive knowledge to advance universal health coverage.

A tale of two exemplars: the maternal and newborn mortality transitions of two state clusters in India.

BACKGROUND: India's progress in reducing maternal and newborn mortality since the 1990s has been exemplary across diverse contexts. This paper examines progress in two state clusters: higher mortality states (HMS) with lower per capita income and lower mortality states (LMS) with higher per capita income. METHODS: We characterised state clusters' progress in five characteristics of a mortality transition model (mortality levels, causes, health intervention coverage/equity, fertility and socioeconomic development) and examined health policy and systems changes. We conducted quantitative trend analyses, and qualitative document review, interviews and discussions with national and state experts. RESULTS: Both clusters reduced maternal and neonatal mortality by over two-thirds and half respectively during 2000-2018. Neonatal deaths declined in HMS most on days 3-27, and in LMS on days 0-2. From 2005 to 2018, HMS improved coverage of antenatal care with contents (ANCq), institutional delivery and postnatal care (PNC) by over three-fold. In LMS, ANCq, institutional delivery and PNC rose by 1.4-fold. C-sections among the poorest increased from 1.5% to 7.1% in HMS and 5.6% to 19.4% in LMS.Fewer high-risk births (to mothers <18 or 36+ years, birth interval <2 years, birth order 3+) contributed 15% and 6% to neonatal mortality decline in HMS and LMS, respectively. Socioeconomic development improved in both clusters between 2005 and 2021; HMS saw more rapid increases than LMS in women's literacy (1.5-fold), household electricity (by 2-fold), improved sanitation (3.2-fold) and telephone access (6-fold).India's National (Rural) Health Mission's financial and administrative flexibility allowed states to tailor health system reforms. HMS expanded public health resources and financial schemes, while LMS further improved care at hospitals and among the poorest. CONCLUSION: Two state clusters in India progressed in different mortality transitions, with efforts to maximise coverage at increasingly advanced levels of healthcare, alongside socioeconomic improvements. The transition model characterises progress and guides further advances in maternal and newborn survival.

Messy but worth it: human-centred design as applied within a successful vaccine-promotive campaign.

Human-centred design (HCD) is an approach to problem-solving that prioritises understanding and meeting the needs of the end-users. Researchers and designers practice empathic listening as users share their perspectives, thereby enabling a variety of stakeholders to cocreate effective solutions. While a valuable and, in theory, straightforward process, HCD in practice can be chaotic: Practitioners often struggle to navigate an excess of (often conflicting) ideas and to strike a balance between problem-understanding and problem-solving. In this practice paper, we outline our own experiences with HCD, which ultimately resulted in the development of a successful video-based intervention to bolster vaccine confidence in the Philippines. We highlight the use of 'radical circles' to overcome roadblocks and navigate tensions. Radical circles entail groups of individuals with divergent opinions and identities engaging in critical analysis of a given idea, actively challenging standard ways of thinking, and ultimately, generating solutions. Employing radical circles enabled us to innovate and adapt to new perspectives that emerged along the non-linear HCD pathway. Our incorporation of radical circles into HCD methodology demonstrates its potential as a powerful complementary step in the meaning-making process. In our view, radical circles could enrich HCD processes and provide a solution to design overcrowding, leading to meaningful, transformative and successful interventions.

Translating the consent form is the tip of the iceberg: using cognitive interviews to assess the barriers to informed consent in South African health facilities.

The increasing digitisation of personal health data has led to an increase in the demand for onward health data. This study sought to develop local language scripts for use in public sector maternity clinics to capture informed consent for onward health data use. The script considered five possible health data uses: 1. Sending of general health information content via mobile phones; 2. Delivery of personalised health information via mobile phones; 3. Use of women's anonymised health data; 4. Use of child's anonymised health data; and 5. Use of data for recontact. Qualitative interviews (n = 54) were conducted among women attending maternity services in three public health facilities in Gauteng and Western Cape, South Africa. Using cognitive interviewing techniques, interviews sought to:(1) explore understanding of the consent script in five South African languages, (2) assess women's understanding of what they were consenting to, and (3) improve the consent script. Multiple rounds of interviews were conducted, each followed by revisions to the consent script, until saturation was reached, and no additional cognitive failures identified. Cognitive failures were a result of: (1) words and phrases that did not translate easily in some languages, (2) cognitive mismatches that arose as a result of different world views and contexts, (3) linguistic gaps, and (4) asymmetrical power relations that influence how consent is understood and interpreted. Study activities resulted in the development of an informed consent script for onward health data use in five South African languages for use in maternity clinics.

In the wake of growing digitisation of personal health data, greater scrutiny is needed on the language of informed consent and the processes for soliciting consent in health care facilities. Qualitative interviews using cognitive interviewing techniques were used to develop and refine consent language in English, Sesotho, isiXhosa, isiZulu and Setswana for the onward use of health data among maternity clients in public sector primary health clinics in the Western Cape and Gauteng provinces of South Africa. We found that translation in local languages and addressing individual words and phrases was only one barrier to requesting informed consent. Other barriers were cognitive mismatches between the question intent and how women understood the question, linguistic gaps that were linked to language and identity, and power dynamics that affected how women understood the consent script. Emerging language scripts used "/" to present words in multiple languages; a reflection of the multi-linguistic nature of communities in this context.