Grounded Theory

Grounded Theory in Health Economics and HTA

Grounded theory starts from data rather than from a hypothesis, with researchers analysing each interview or document as it arrives and choosing whom to sample next from what the analysis has found so far. Health economists meet it mainly at the front end of quantitative work, where qualitative interviews decide which attributes a preference study or a capability measure should contain and how people understand the wording. It also produces the kind of evidence on patient experience that health technology assessment committees weigh, and the NICE manual (PMG36) names formal qualitative research studies as one route by which such evidence arrives. Many studies in the field borrow single grounded theory techniques, above all constant comparison, without building a full theory, and that difference matters when the work is appraised.

Origins in a study of dying patients

Glaser and Strauss developed the approach while studying terminally ill patients who had differing knowledge of their health status, work first reported in Awareness of Dying in 1965. During that study they developed the constant comparative method, and their 1967 book The Discovery of Grounded Theory explained how theory could be generated from data inductively instead of being tested deductively.

The founders later diverged, and Glaser wrote a rebuttal of Strauss and Corbin's Basics of Qualitative Research. Chun Tie and colleagues distinguish three genres.

The genres differ in the researcher's philosophical position, in the use of literature, and in the approach to coding and theory development. A constructivist researcher co-constructs experience and meanings with participants, and Birks and Mills, as summarised by Chun Tie and colleagues, hold that theory is constructed by the researcher instead of being discovered.

How a grounded theory study proceeds

A study usually opens with purposive sampling of participants who can answer the research question. Collection and analysis then run together, so the first transcripts are coded before further data are gathered, and the coding shapes the next round of questions and recruits.

Constant comparative analysis drives the work. Incidents in the data are coded and compared with each other, codes are compared with other codes and collapsed into categories, and new data are compared with what was found earlier. Chun Tie and colleagues describe this continual comparison as what sets grounded theory apart from purely descriptive analysis.

Theoretical sampling follows leads from the analysis by recruiting new participants or material that can fill gaps in a developing category. It serves the development of the theory, which is why a grounded theory sample is not built for population representation in the way a random sampling design is. Memos record the analyst's decisions and form an audit trail.

Coding moves from labelling small segments, often in the participants' own words, towards a core category that organises the others. The analysis ends with theoretical saturation, when new data add nothing to the existing categories and those categories are sufficiently explained. The result is a substantive theory, presented as a set of interrelated concepts with explanatory statements linking them to the core category, as opposed to a list of themes. Under Glaser's dictum that all is data, documents and focus group transcripts can feed the analysis, although interviews are the most common source.

Saturation and sample size

Saunders and colleagues trace the idea of saturation to grounded theory, where Glaser and Strauss tied it to sampling for a category until no additional data develop its properties. They identify four models in later use, theoretical saturation, inductive thematic saturation, a priori thematic saturation and data saturation, which differ in whether the stopping rule concerns the theory, the emergence of new codes, the representation of pre-set codes, or simple repetition in the data.

They also cite Mason's observation that interview sample sizes are commonly multiples of ten, a sign that rules of thumb often replace an adaptive stopping rule. Unlike a sample size calculation based on statistical power, a grounded theory sample size is a reported judgement about the analysis, so a protocol should say which kind of saturation it will seek.

Grounded theory methods in outcome and preference research

The ICECAP-A capability measure for adults shows the techniques at work in measure development. Al-Janabi, Flynn and Coast interviewed 36 purposively sampled adults in England and re-interviewed 18 of them to develop wording. Interviewing continued until saturation was achieved, transcripts were analysed in batches of five to eight, and constant comparative methods citing Glaser's 1965 paper were used to compare extracts across informants and against emerging themes.

The study produced five attributes (stability, attachment, achievement, autonomy and enjoyment), each with one item and four response categories. The needs of economic evaluation shaped that result, because valuation exercises can only cope with relatively small numbers of attributes and levels. Living in a just world mattered to some informants but was rejected, partly because it could dominate the other attributes in valuation.

Not every ICECAP study used these methods. The earlier older people's measure (Grewal and colleagues, 40 informants aged 65 and over) analysed its interviews with Framework analysis.

Discrete choice experiments show the same pattern. Vass, Rigby and Payne reviewed 254 healthcare DCEs, of which 29 reported or cited extensive qualitative work. Within those 29, thematic analysis was the most commonly stated approach (10 studies) and five reported constant comparative analysis. Coast and Horrocks, developing attributes for dermatology services through interviews with 19 patients, described a clear tension between the qualitative aim to explore and describe and the reductiveness needed to encapsulate a service in a minimum number of attributes.

A full grounded theory study aimed at HTA

Szabo, Hawkins and Germeni used a complete grounded theory design to study why patients with moderate to severe plaque psoriasis prefer different treatments. They interviewed 38 people in Canada between October 2023 and December 2024, guided theoretical sampling by treatment history, and coded transcripts with open, axial and selective coding alongside memos. As the analysis developed they added minimum quotas of three people with experience of newer oral drugs, three who had stopped biologic therapy, and three without employer-sponsored drug coverage.

The core category was heterogeneous treatment preferences. Treatment priorities fell under two headings, controlling symptoms or minimising the negative impact of treatment, and every participant valued effectiveness. Convenience meant a daily pill to some and a monthly injection to others.

The authors noted that such heterogeneity is not typically considered in HTA decisions, and they cautioned that, because theoretical sampling guided recruitment, the frequencies of patient characteristics in the sample are not intended to reflect the wider population with psoriasis.

Where qualitative evidence enters an appraisal

The NICE manual (PMG36) states that qualitative research can explore values, preferences, acceptability, feasibility and equity implications, including patients' experience of a condition and of a treatment and the views of carers and clinicians. Qualitative data may come ad hoc, through formal qualitative research studies, or from a systematic review.

When qualitative evidence is extensive, PMG36 prefers recognised methods of analysis and synthesis and gives rapid review, framework synthesis, narrative summary and synthesis, meta-synthesis and thematic synthesis as examples. The manual does not name grounded theory, so on this reading a grounded theory study enters a NICE health technology assessment as one formal qualitative study among others. In committee decision making, the manual says the committee considers qualitative evidence on the experiences of patients, carers and experts alongside randomised and non-randomised studies.

Qualitative evidence also has a defined role when the reference case measure is challenged. To argue that the EQ-5D is inappropriate, PMG36 asks for qualitative empirical evidence on its lack of content validity, showing that key dimensions of health are missing, supported by evidence of poor construct validity and responsiveness.

Worked example of saturation tracking in an attribute study

The figures below are illustrative. An analyst developing attributes for a patient-reported outcome measure codes interviews in batches of six and records how many codes in each batch are new.

$$s_b = \frac{n_b}{N_b}$$

where $s_b$ is the share of all codes so far that first appeared in batch $b$, $n_b$ is the number of new codes in batch $b$, and $N_b$ is the cumulative number of codes after batch $b$.

For batch 4 the calculation is shown below.

$$s_4 = \frac{7}{56} = 0.125$$

where 7 is the number of new codes in batch 4 and 56 is the cumulative count after it.

A fixed rule to stop once fewer than 10% of codes were new would have ended recruitment after batch 3 at 18 interviews. The analysis at that point suggested that carers' accounts could change one category, and sampling them added seven codes. The count of new codes tracks what Saunders and colleagues call inductive thematic saturation, while grounded theory stops on theoretical saturation, a judgement that new data add nothing to the categories and each is sufficiently explained. The 57 codes must then be condensed into a few attributes, where Coast and Horrocks located the tension between description and reduction.

Boundaries with nearby methods

Thematic and Framework analysis organise qualitative data into themes and are common in health economics, as the ICECAP older people's study and the DCE review show. Grounded theory aims further, at an explanatory theory built round a core category, with codes that identify processes and actions.

A focus group is a data collection method that a grounded theory study may use, and the Delphi method seeks expert agreement, a different aim. A grounded theory study usually collects new data as primary research, although Chun Tie and colleagues also list documents and government reports as sources, while synthesis methods listed in PMG36, such as meta-synthesis and thematic synthesis, combine findings across existing studies.

Where grounded theory claims go wrong

The commonest misunderstanding is to call any inductive coding grounded theory. Chun Tie and colleagues present the method as interlocking parts, with concurrent collection and analysis, constant comparison, theoretical sampling, memoing and theoretical saturation. A study that applies constant comparison without theoretical sampling or a theory-building aim uses grounded theory methods, in the phrase Vass and colleagues apply to one DCE study, without being a grounded theory study.

A second error is to read counts from a grounded theory sample as prevalence. Theoretical sampling deliberately over-recruits informative cases, so the share of participants holding a view should not be carried into a model as a probability or used to weight subgroups in a cost-effectiveness analysis.

Researcher influence is a recognised weakness. Al-Janabi and colleagues noted that different researchers might have produced different attributes, and Coast, McDonald and Baker identified generalisability and reflexivity as issues affecting the acceptability of qualitative methods. Coast warned in 1999 that health economists who use qualitative methods naively risk producing research accepted neither by economists nor by qualitative researchers.

Sources

• Chun Tie Y, Birks M, Francis K. Grounded theory research: a design framework for novice researchers. SAGE Open Medicine 2019;7:2050312118822927. https://doi.org/10.1177/2050312118822927
• Saunders B, Sim J, Kingstone T, et al. Saturation in qualitative research: exploring its conceptualization and operationalization. Quality and Quantity 2018;52(4):1893-1907. https://doi.org/10.1007/s11135-017-0574-8
• Coast J. The appropriate uses of qualitative methods in health economics. Health Economics 1999;8(4):345-353. https://doi.org/10.1002/(SICI)1099-1050(199906)8:4<345::AID-HEC432>3.0.CO;2-Q
• Coast J, McDonald R, Baker R. Issues arising from the use of qualitative methods in health economics. Journal of Health Services Research and Policy 2004;9(3):171-176. https://doi.org/10.1258/1355819041403286
• Al-Janabi H, Flynn TN, Coast J. Development of a self-report measure of capability wellbeing for adults: the ICECAP-A. Quality of Life Research 2012;21(1):167-176. https://doi.org/10.1007/s11136-011-9927-2
• Grewal I, Lewis J, Flynn T, Brown J, Bond J, Coast J. Developing attributes for a generic quality of life measure for older people: preferences or capabilities? Social Science and Medicine 2006;62(8):1891-1901. https://doi.org/10.1016/j.socscimed.2005.08.023
• Vass C, Rigby D, Payne K. The role of qualitative research methods in discrete choice experiments. Medical Decision Making 2017;37(3):298-313. https://doi.org/10.1177/0272989X16683934
• Coast J, Horrocks S. Developing attributes and levels for discrete choice experiments using qualitative methods. Journal of Health Services Research and Policy 2007;12(1):25-30. https://doi.org/10.1258/135581907779497602
• Szabo SM, Hawkins NS, Germeni E. Controlling symptoms versus minimizing negative treatment impact: a grounded theory study of heterogeneity in patient preferences for psoriasis treatments. Patient Preference and Adherence 2026;20:604216. https://doi.org/10.2147/PPA.S604216
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. https://www.nice.org.uk/process/pmg36