A cross sectional, observational study was conducted of all Aotearoa New Zealand general practices and enrolled patients as of 30 September 2018. Māori investigators provided governance to the project with respect to selecting outcomes and explanatory variables and interpreting results. What follows is a brief description of methods; further details can be found in the primary outcomes report from this study [22].
Data sources
Data came from national datasets, held by the Ministry of Health, and from practice information held by PHOs. National datasets included PHO registers, inpatient, outpatient, laboratories, pharmaceutical dispensing, immunisations, the Virtual Diabetes Register (VDR), NZDep2018 Index of Deprivation, the Index of Multiple Deprivation (IMD) and the Measuring Multimorbidity Index (M3), all available at patient level.
The VDR lists all individuals considered to have diabetes at 31 December 2018 based on linking six national administrative datasets [30]. The NZDep2018 combines nine variables from the 2018 census which reflect eight dimensions of material deprivation, assigning a score to each resident in a small geographic area [32]. The IMD assigns a deprivation score to residents of a larger geographic area. The index is constructed from seven domains, which can be used independently: employment, income, crime, housing, education, health, and access. We used all domains except health [33]. The M3 index is a score assigned to each individual based on number and type of conditions they have, derived from hospital discharge coding [34]. Data from these sources were available for all practices and patients.
The national PHO register lists patients enrolled in each practice. A patient unique identifier, their National Health Index (NHI), is used throughout the health system. Processes linking patient-level data using an encrypted NHI are well-established.
All PHOs extract data from practices electronic records, although details vary between PHOs. Ten PHOs, with 292 practices, contributed patient-level data. Data from appointment books was used to calculate number and length of consultations and the profession of the clinician seen, for face-to-face consultations, but not telephone, email or other contacts.
The practice data also allowed calculation of three items of preventive care: rates of cervical screening, cardiovascular risk assessment and HbA1c testing (which also drew on data from the national laboratory dataset).
The workforce numbers and Full Time Equivalents came from a survey sent to practices by all participating PHOs. Data on GP FTE came from 370 practices, and RN FTE came from 367 practices but not all data were complete and comparable. The FTE calculations were based on data that covered 12% of patients.
Defining practice models
• Traditional practice: Typically centred upon the general practitioner, with mainly nursing support, operating as a small business, and owned by one or more doctors. These ranged from small to large organisations and served both high need and lower need populations. This is the longest-standing model. Individual practices have a high degree of autonomy over service delivery.
• Corporate practice: A group of practices owned and run as a for-profit business entity. Some delivered high volumes of care, with low costs for patients and often without the need for an appointment. Corporate practices had a relatively high degree of standardisation in business and clinical processes and information technology across different sites.
• Health Care Home (HCH): the New Zealand HCH Collaborative maturity matrix focuses on business efficiency and sustainability [18]. The first practice formally enrolling in the programme in 2011. Only 14 had been fully certificated as mature HCHs by 30 September 2018 (A Maxwell, personal communication 2018). At the time of this study those not certificated were at different stages of meeting the maturity matrix criteria.
• PHO/DHB practices: Practices owned by a PHO or a DHB. This was a small group that had mostly been taken over by a PHO or DHB to continue to provide primary care services in a specific location, often an underserved and/or rural area.
• Trust/NGO practices: One or more practices owned by an entity that was a not-for-profit Trust or NGO. They had a stated purpose, identifying a health or social goal. Many were in small communities or served populations with high need. Some provided, for example, salary and premises to attract and retain staff.
• Māori practices: Practices owned and governed by Māori organisations, serving Māori and non-Māori patients. They were identified through lists from the Ministry of Health and DHBs together with web searches, direct contact with practices or known to investigators. There may be a small number of practices we did not identify as Māori practices.
• Pacific practices: Practices owned and governed by Pacific organisations, serving mostly Pacific and some non-Pacific patients. They were identified through lists from the Ministry of Health and DHBs together with web searches, direct contact with practices or known to investigators. There may be a small number of practices we did not identify as Pacific practices.
Traditional, Corporate, PHO/DHB or Trust/NGO were also considered to be ownership types. We assigned every practice to one of these ownership types. HCH, Māori and Pacific practices could overlap with ownership types, and HCH could overlap with Māori and Pacific practices.
Patient health outcomes
Outcome measures were selected from existing performance indicators within collections of the New Zealand Health Quality and Safety Commission or the New Zealand Health Quality Measures collections [35]. Measures were known to show significant inequities between groups by health need, material deprivation or ethnicity but none had previously been examined for variation by primary care model of care. The six study outcomes used were as follows.
• Polypharmacy: Patients taking 5 or more long term medications over two consecutive quarters [36]. For this paper, where regressions include only Māori patients, polypharmacy was calculated on patients aged 55 years and over.
• HbA1c Testing: Patients on the national VDR with one or more HbA1c test in the previous year.
• 6 Month Immunisation: Children who had received, by age 6 months, all the scheduled childhood immunisations up to and including those due at 5 months. The calculation includes only children who were 6 months old at some point during the analysis period. The Ministry of Health definition of on-time immunisation allows for a window of 1 month after the due date [37].
• Child ASH Admissions: The number of ambulatory sensitive hospital admissions for children who were under 15 years of age at the end of the analysis period [38, 39].
• Adult ASH Admissions: The number of ambulatory sensitive hospital admissions for adults who were between 45 and 64 years of age at the end of the analysis period [39].
• ED Attendances: The number of attendances at an ED for each patient over the analysis period.
The analysis period was the year 1 October 2017 to 30 September 2018. All measures used the national data sets. No adjustment was made for reduced data from those who died during the year of observation. Three outcomes were process measures: polypharmacy, HbA1c testing and childhood immunisations. Three were measures of intermediate outcomes: child and adult ASH and ED attendances. Better outcomes were assumed to be lower polypharmacy, ASH and ED attendances, and higher HbA1c testing and childhood immunisations.
Explanatory variables
Patient characteristics. Patients were assigned to the practice in which they were registered in the national PHO database at 30 September 2018. Age, gender and ethnicity were available at that date. Living in a deprivation area quintile 5 (most deprived quintile, Q5), Index of Multiple Deprivation (IMD) score of the area the patient lives in, distance to the nearest ED, the M3 score and being on the VDR all used 2018 data. Having gout was determined from dispensing data back to 2001 and hospital discharge data back to 1988 [40]. Having gout and diabetes are both associated with other long term conditions [41]. Being dispensed a selective serotonin reuptake inhibitor (SSRI, usually for depression), dispensed tramadol (for moderate to severe pain), dispensed an antibiotic, patient changing enrolled practice (a measure of practice continuity), and number of first medical specialist assessment (FSA) attended and not attend were all measured in the year 1 October 2017 to 30 September 2018.
Practice characteristics. VLCA practices agree to receive increased capitation funding while limiting their fees to patients. Practice uptake of this contract is voluntary subject to having an enrolled population of ≥ 50% Māori, Pacific or people living in Quintile 5 areas. Practices were designated as either urban or rural based on the rural status of a majority of their enrolled patients. The percentage of patient consultations, in the previous year, with the same GP, was used as a measure of personal continuity.
Primary care clinical input. Face-to-face appointments, recorded in the practice appointment record, were attributed to a RN, NP, GP or Other. Total Consultations refers to the number of consultations recorded in the appointment book with a GP or NP in the previous year. There were low numbers of NPs and NP consultations so we made a decision to combine with GP. Low numbers of NPs would have been difficult to interpret. By combining with GP we were able to see more clearly the effect on patient outcomes of an independent consultation which a GP and NP did routinely. Time spent with each patient was cumulated to a proportion of Full Time Equivalent (FTE) per 1000 enrolled patient, separately for GPs, NPs and RNs, where that information was available.
Regression analyses
Multilevel mixed effects regression analyses used (only Māori) patient-level data adjusted for clustering at practice level. All analyses were conducted in R statistical software [42, 43]. Variables that do not appear in the final regressions were not statistically significant in development models. The comparators used in the regressions vary between practice models. Model of care categories Corporate, PHO/DHB and Trust/NGO were compared to Traditional. HCH, Māori practices and Pacific practices were compared with not-HCH, not-Māori practices and not-Pacific practices, respectively. Variance was partitioned at the level of patient and practice, but not at level of model of care because a given practice might be classified to more than one model.
Practice model and living in deprivation were entered as independent explanatory variables, and interpreted to imply inequity if there was a significant association between these variables and patient health outcome. Statistical significance is cited at p≤0.05, with no adjustment for repeated modelling and multiple outcomes. Estimations for child ASH, adult ASH and ED attendances were run on 50%, 25% and 12.5% samples of the data respectively, to avoid excessively long computations with the negative binomial regressions used for count data.