Inequities in primary care respiratory illness presentations in Aotearoa (New Zealand) before, during and after the COVID-19 pandemic: a retrospective cohort study using automated health record analysis Authors: Nikki Turner, Anthony Dowell, Rachel Blanch, Mika Hiroi, Lorraine Castelino, Jayden MacRae Published: 2026 ABSTRACT AbstractBackgroundEthnic and socioeconomic inequities in respiratory illness outcomes are well established. Monitoring primary care presentation rates is important for assessing health system equitable responsiveness, particularly during health crises.AimTo examine changes in primary care respiratory illness presentation rates in Aotearoa New Zealand (NZ) by ethnicity and socioeconomic status before, during and after the COVID-19 pandemic.Design and settingRetrospective cohort study using natural language processing (NLP) of electronic health records from 37 general practices across eight districts in NZ.MethodClinical interactions from 2018 to 2023 (n=10 421 399; 4 66 349 patients) were classified into respiratory categories using a validated NLP algorithm. Age-adjusted monthly presentation rates were analysed by ethnicity and deprivation across pre-pandemic, pandemic and post-pandemic periods.ResultsOverall, 15.1% (1 573 631/10 421 399) of interactions were respiratory-related, increasing from 13.7% to 14.2% in 2018–2019 to a peak of 25.2% in 2022, before declining to 15.3% in 2023. Pacific peoples had higher presentation rates than European patients both pre-pandemic (+2.2, p=0.03) and during the pandemic (+3.5, p<0.01), and the presentation rates for Pacific peoples more than doubled during the pandemic peak. Post-pandemic, rates declined across all groups with larger reductions for Pacific peoples. Higher presentation rates were consistently observed among those in the most deprived quintile. Acute respiratory presentations varied more over time, while chronic conditions remained relatively stable.ConclusionsPrimary care presentation rates increased during the pandemic, particularly among high-needs populations, suggesting responsiveness to need. Post-pandemic declines in acute presentations, especially among Pacific peoples, may indicate emerging access barriers and widening inequities. NLP applied to routinely collected electronic health record data provides a retrospective approach to analysing primary care presentations and may support future development of surveillance approaches for monitoring equity trends. STRENGTHS AND LIMITATIONS OF THIS STUDY This methodology allows the measurement over time of important equity issues, including socioeconomic and ethnic measures. This methodology allows the measurement over time of important equity issues, including socioeconomic and ethnic measures. Natural language processing (NLP) methodology can add important information to primary care for policy and service planning during times of health crisis. Natural language processing (NLP) methodology can add important information to primary care for policy and service planning during times of health crisis. NLP methodology relies on routine general practice consultations rather than a research base, so it may be affected by consultation styles over time. NLP methodology relies on routine general practice consultations rather than a research base, so it may be affected by consultation styles over time. This methodology does not assess actual disease rates or access; it measures realised access. This methodology does not assess actual disease rates or access; it measures realised access. Introduction Globally, inequities in access to healthcare and health outcomes remain persistent and well-documented, particularly in relation to the social determinants of health.13 In Aotearoa New Zealand (NZ), Pacific peoples, Māori (indigenous NZ) and individuals living in areas of greater socioeconomic deprivation experience disproportionately higher rates and severity of respiratory illness.24 These inequities are influenced by systemic factors, including reduced access to healthcare, delayed diagnosis and overcrowded housing. Internationally, the COVID-19 pandemic further amplified longstanding health inequities, with disproportionately adverse outcomes reported among ethnic minority and socioeconomically disadvantaged populations.5 NZ implemented an early and intensive public health response to the pandemic, including international border restrictions and national lockdowns. These measures substantially reduced or eliminated community transmission of other respiratory viruses, particularly influenza and respiratory syncytial virus.6 Auckland, the largest urban area in NZ, experienced the earliest and the most substantial impact from the pandemic. Studies conducted during the pandemic reported changes in primary care presentation rates and inequities in healthcare access by ethnicity and socioeconomic status.7 8 However, most research has focused on the acute pandemic period, with limited evidence examining post-pandemic respiratory presentation patterns in primary care. Understanding how respiratory consultation patterns changed before, during and after the pandemic is important for informing future health system preparedness and equitable service planning. Prior to the pandemic, Pacific peoples and Māori consistently experienced higher rates and greater severity of respiratory illness.9 10 International evidence also suggests that the pandemic differentially affected ethnic groups in accessing consultations.11 12 However, there remains limited evidence describing longer-term changes in primary care respiratory presentations across population groups post-pandemic. Primary care electronic health records (EHRs) contain rich clinical information that can provide valuable insights into healthcare utilisation and disease presentation patterns. However, these data are often difficult to access and analyse for research or surveillance purposes, particularly where clinically relevant information is embedded within unstructured consultation notes. A natural language processing (NLP) programme has been validated and used in NZ primary care to investigate presentation patterns across a range of clinical conditions.1315 More recently, international EHR data have been used successfully to describe the impact of COVID-19 in primary care.16 By incorporating both structured and unstructured clinical data, NLP methods enable more comprehensive analysis of respiratory presentations within primary care populations. The aim of this study was to examine changes in presentation rates in NZ general practice before, during and after the COVID-19 pandemic, with a particular focus on differences by ethnicity, socioeconomic deprivation and geographic region. Methods Study design This study used a retrospective observational design applying a validated NLP algorithm to analyse routinely collected primary care EHR data. The approach incorporated both quantitative and qualitative clinical information derived from structured and unstructured consultation records. This methodology was a rule-based classification system that had been validated and published previously.13 14 Patient interactions recorded in general practice EHRs were classified using a text-based clinical classification system designed to identify respiratory presentations based on documented clinical characteristics. Rates, volumes and temporal variation in respiratory interactions were analysed across the predefined periods: pre-pandemic (January 2018–February 2020), pandemic (March 2020–December 2022) and post-pandemic (January 2023–December 2023). A supplementary stratified analysis comparing Auckland with the remainder of NZ was undertaken to explore geographic variation in respiratory presentation patterns, given differences in population composition and the disproportionate impact of COVID-19 outbreaks and public health restrictions within Auckland. This study uses the same underlying dataset and analytical framework as a related study examining overall changes in respiratory illness presentation patterns during the COVID-19 pandemic. However, this analysis addressed a distinct research question by focusing specifically on inequities in respiratory presentation rates according to ethnicity and socioeconomic deprivation. Setting and participants All NZ residents and citizens are assigned a unique National Health Index (NHI) number and are enrolled with a primary care practice.15 The NHI number enables linkage of healthcare interactions across both primary and secondary care settings and facilitates integration with demographic information contained within EHRs, including ethnicity and socioeconomic deprivation. Individuals may report up to six ethnicities within the NHI system. For analysis, prioritised ethnicity classification was applied to ensure each individual contributed to a single ethnicity category. Ethnicity data were aggregated into five categories: Māori, Pacific, Asian, European and other.17 Socioeconomic deprivation was determined using the NZDep2018 Index of Deprivation, based on patient domicile at the meshblock level, the smallest census geographic unit in NZ.18 NZDep2018 scores incorporate measures including income, employment, education, housing and access to transport. Scores were categorised into quintiles, ranging from 1 (least deprived) to 5 (most deprived). Both ethnicity and deprivation were mutable for any given patient. To ensure consistent categorisation across longitudinal analysis, patients were assigned their most recent prioritised ethnicity and deprivation quintile recorded within the study period. Data were obtained from consenting general practices across eight districts in NZ and included both urban and rural populations. The dataset comprised all recorded general practice interactions between 2018 and 2023. No identifiable patient information was included in study outputs. Consent for the use of de-identified routinely collected data was incorporated within standard patient enrolment processes at participating general practices. Data processing and NLP classification A validated NLP algorithm was used to analyse clinical consultation records from 2018 to the end of 2023. The system, previously described in earlier publications,1315 classifies EHR data using a combination of structured information (including Read codes for diagnoses and prescribing data) and unstructured clinical narrative from general practitioner and practice nurse consultations. The analytical workflow followed a structured classification pipeline (see figure 1). As previously reported,11 first, all consultation records were classified by the NLP algorithm into two primary groups: respiratory and non-respiratory consultations. Respiratory consultations included conditions affecting the respiratory tract, including otitis media, while non-respiratory consultations included all other presentations. Screening-only consultations were excluded to avoid overestimation of respiratory presentation burden. Analytical workflow of the natural language processing (NLP) pipeline and statistical analysis. EHR, electronic health record. Respiratory consultations were subsequently classified by the NLP algorithm into nine predefined diagnostic categories11: (1) asthma, (2) chronic obstructive pulmonary disease(COPD), (3) influenza-like illness, (4) lower respiratory tract infection (LRTI), (5) non-upper respiratory tract infection rhinitis, (6) otitis media, (7) sore throat, (8) upper respiratory tract infection (URTI) and (9) wheeze illness. COVID-19-related consultations were additionally classified as either confirmed COVID-19 positive (COVID-POS) or COVID-19 discussed/negative/unspecified. The NLP algorithm was developed and validated using a supervised training approach. Two general practice clinicians independently annotated 6575 consultation records used for model development. Performance was evaluated using an independent hold-out validation dataset of 3277 consultation records. Bootstrap resampling (1000 iterations) was used to generate 95% CIs for positive predictive value, sensitivity and specificity across all diagnostic categories (online supplemental table S1). Additional rule-based logic was subsequently applied to refine respiratory subcategories. Wheeze, asthma and chronic lung disease presentations were initially grouped before reclassification according to age and smoking history. Children under 5 years were classified as wheeze illness, while individuals aged 5 years and older were classified as asthma unless COPD or other chronic lung disease codes were present or the patient was over 50 years with a recorded smoking history. Statistical analysis Classified consultation data were aggregated into three time periods: pre-pandemic (2018–early 2020), pandemic (early 2020–end 2022) and post-pandemic (2023). Monthly age-standardised respiratory presentation rates were calculated and stratified by ethnicity, socioeconomic deprivation quintile and respiratory diagnostic category. Generalised least squares regression was used to account for temporal autocorrelation in longitudinal trends. CIs for rate estimates across demographic strata were generated using bootstrap resampling (1000 iterations). Results Study population characteristics A total of 466 349 individual patients contributed 10 421 399 patient interactions between 2018 and 2023 across 37 participating practices. Baseline demographic characteristics of the study cohort are presented in table 1. The cohort included patients across all age groups, ethnicities, gender categories and socioeconomic deprivation quintiles, reflecting a broad cross-section of individuals attending participating general practices during the study period. Table 1Baseline characteristics of the study cohortDemographicCategoryStudy cohort (%)EthnicityAsian16.0European53.5Māori14.3Other8.7Pacific7.5GenderMale48.9Female50.5Other0.5Age group (years)0–410.35–94.910–145.115–196.020–247.825–299.430–348.735–397.140–446.245–496.350–546.055–595.660–644.865–694.070–742.975–792.180–841.385+1.2Deprivation quintileUnknown5.81 (least deprived)18.3218.7318.7419.05 (most deprived)19.5 Baseline characteristics of the study cohort Comparison with the NZ census population is provided in online supplemental table S2. Relative to the national population, the study cohort included a higher proportion of Asian and Pacific peoples, a lower proportion of individuals of European ethnicity and a slightly younger age distribution. Overall respiratory presentation trends Respiratory presentations accounted for 15.1% (1 573 631) of the 10 421 399 recorded general practice interactions between 2018 and 2023. The proportion of respiratory presentations remained relatively stable between 2018 and 2019 (13.7–14.2%), increased during the pandemic period (15.5% in 2020 and 16.0% in 2021), peaked at 25.2% in 2022 and declined to 15.3% in 2023. These temporal changes in respiratory presentation rates provide the context for subsequent analyses by ethnicity, socioeconomic status and geographic region. Respiratory presentation patterns by ethnicity Figure 2 shows consistent overall consultation volumes over time across ethnic groups, while respiratory presentation patterns showed marked seasonal disruption during periods of SARS-CoV-2 transmission and public health restrictions. Generalised least squares regression analysis demonstrated that Pacific peoples had the highest respiratory presentation rates prior to the pandemic (+2.2 compared with European populations, p=0.03) and throughout the pandemic period (+3.5 compared with European populations, p<0.01). Monthly age-adjusted respiratory interaction rates (%) for Māori, Pacific, Asian, NZ European and ‘Other’ ethnic groups; shaded areas indicate COVID-19 Alert Level and Protection Framework (PF) periods (level 2/orange, level 3/red, level 4 most restrictive) Following the emergence of the SARS-CoV-2 Delta outbreak in 2021, respiratory presentations among Pacific peoples increased substantially, rising from approximately 20% to 45% of consultations. Māori presentations also increased during this period, from under 20% to approximately 35%. Post-pandemic respiratory presentation rates remained elevated among Pacific peoples for approximately 8 months before declining towards rates comparable with those of other ethnic groups by late 2023. Compared with the pre-pandemic period, post-pandemic respiratory presentation patterns demonstrated lower winter peaks and higher summer troughs across all ethnic groups, with Pacific peoples showing the greatest reduction in peak seasonal respiratory presentations. Respiratory presentation patterns varied across ethnic groups and diagnostic categories over time, with greater temporal variation observed for acute respiratory conditions than chronic respiratory disease presentations, figure 3. For URTIs, Asian populations demonstrated higher presentation rates prior to the COVID-19 pandemic. During the COVID-19 period, higher presentation rates were observed among Pacific peoples, but post-pandemic Asian presentations returned to the pre-COVID pattern. For LRTIs, Pacific peoples consistently demonstrated the highest presentation rates prior to the pandemic, followed by a greater relative reduction in post-pandemic LRTI presentations compared with other ethnic groups. Māori experienced consistently higher chronic lung disease presentation rates throughout the study period, followed by Pacific peoples. Monthly age-adjusted respiratory interactions (%) by respiratory group category, including asthma, chronic lung disease, COVID-19, ILI, lower RTI, rhinitis, otitis media, sore throat, URTI and wheeze. Estimates are stratified by ethnicity (Māori, Pacific, Asian, European and ‘Other’). COVID-19-NOS, COVID-19 discussed/negative/unspecified; ILI, influenza-like illnInfluenza-like Illness; RTI, respiratory tract infection; URTI, upper RTI. Asthma-related presentations increased markedly immediately prior to the first national lockdown in 2020. COVID-19-related presentations were also more pronounced among Pacific peoples compared with other ethnic groups during the pandemic period. Geographic variation in respiratory presentation patterns A supplementary stratified analysis comparing Auckland with the remainder of NZ is presented in figure 4 to provide contextual information on geographic variation in respiratory presentation patterns during the study period. Prior to the pandemic, respiratory presentation rates were broadly similar across regions. During the pandemic period, Auckland experienced substantially higher respiratory presentation rates compared with the remainder of NZ. A marked increase in COVID-19-related presentations occurred immediately prior to the emergence of the SARS-CoV-2 Delta variant in August 2021. This increase was most pronounced among Pacific peoples in Auckland but was also observed among Māori populations. The increase was driven by both confirmed COVID-POS consultations and consultations related to COVID-19 discussion, testing or unspecified COVID-19 concerns. Monthly age-adjusted respiratory interaction rates (%) are shown for Māori, Pacific and combined all other ethnic groups, stratified by region (Auckland region and all other regions in NZ). Shaded areas indicate periods corresponding to COVID-19 Alert Level and Protection Framework (PF) settings (level 2/orange, level 3/red, and level 4 (most restrictive)) Respiratory presentation patterns by socioeconomic deprivation Respiratory presentation rates varied across socioeconomic deprivation quintiles over time (figure 5). In general, individuals in areas of greater socioeconomic deprivation demonstrated higher respiratory presentation rates than those living in less deprived areas. This socioeconomic gradient was most pronounced for chronic lung disease presentations, whereas differences between deprivation quintiles were less marked for URTIs. The magnitude of variation between deprivation groups differed across respiratory diagnostic categories and across time periods. Monthly age-adjusted general practice respiratory interactions (%) by respiratory categories, stratified by deprivation quintile (1=least deprived to 5=most deprived), including overall and specific illness categories. Shaded areas indicate COVID-19 Alert Level and Protection Framework (PF) periods (level 2/orange, level 3/red, level 4 most restrictive) Discussion Main findings This study identified substantial differences in respiratory presentation patterns to NZ primary care by ethnicity and socioeconomic deprivation across the pre-pandemic, pandemic and post-pandemic periods. Pacific peoples, Māori and individuals living in areas of greater socioeconomic deprivation consistently experienced higher respiratory presentation rates than other population groups. These differences were most pronounced during periods of high SARS-CoV-2 transmission, particularly among Pacific peoples in Auckland during the Delta outbreak. Respiratory presentation patterns also differed according to illness category. Acute respiratory conditions demonstrated greater temporal variability throughout the pandemic period, whereas chronic respiratory disease presentations remained comparatively stable. Following the pandemic period, respiratory presentation rates declined across all ethnic groups, with more pronounced reductions observed among Pacific peoples and for LRTIs. Overall, general practice maintained relatively stable respiratory consultation volumes throughout the pandemic period despite major disruptions to healthcare delivery and public health restrictions. However, post-pandemic reductions in acute respiratory presentations may indicate changing healthcare utilisation patterns and potential emerging barriers to primary care access. Geographic and socioeconomic interpretation The findings of this study align with longstanding evidence that primary care presentation rates and experiences vary by ethnic and socioeconomic groups.9 19 20 In the UK, ethnic minority groups reported less positive experiences with primary care9 19 but higher consultation rates.20 Ethnic differences are also observed in the nature of general practitioner consultations, including location and prescription outcomes.21 In this study, prior to the pandemic, respiratory presentations followed a relatively stable seasonal pattern, with consistently higher rates observed among Pacific peoples and Māori. These findings are consistent with previous reports demonstrating higher rates and greater severity of acute respiratory illness among these populations, including during the COVID-19 pandemic.22 23 During the pandemic peak, respiratory presentation rates among Pacific peoples more than doubled, with substantial increases also observed among Māori populations. Unlike some international studies reporting reduced healthcare attendance among ethnic minority groups during the pandemic,24 the present study demonstrated increased primary care respiratory presentations among higher-needs populations during periods of greatest disease burden. This may reflect a responsive primary care system during the acute pandemic phase, with healthcare utilisation increasing in parallel with clinical need. Studies have also demonstrated minimal change for chronic multimorbidity by ethnicity, which aligns with the findings of this study, indicating stability with chronic disease presentations.24 Acute respiratory issues appeared to be more sensitive than chronic conditions to changes in primary care presentation during crises. The sudden peak seen in this data at the start of the pandemic, specific to asthma presentations across the country, is likely to align with the need to access medication just prior to entering a national lockdown. Auckland experienced the highest respiratory presentation rates during the pandemic period, particularly among Pacific peoples.22 As the largest urban centre in NZ and the region most affected by COVID-19 outbreaks and restrictions, Auckland experienced earlier and more intense community transmission.22 Increased respiratory presentations prior to widespread SARS-CoV-2 circulation may also reflect heightened health awareness, testing behaviour and healthcare engagement among communities perceived to be at the highest risk. Post-pandemic, presentation rates declined across all ethnic groups, with larger reductions observed for Pacific peoples and for acute respiratory conditions such as LRTIs. Several explanations may contribute to this pattern. Community behaviour surrounding management of respiratory illness may have changed following the pandemic, with increased self-management and altered thresholds for healthcare seeking.25 26 However, the more substantial reductions observed for acute rather than chronic respiratory presentations may also suggest emerging barriers to primary care access following the pandemic period. These findings are important within the broader context of increasing strain on NZ primary care services following the pandemic, including workforce shortages, reduced enrolment capacity and longer waiting times.27 28 Greater reductions in respiratory presentations among populations with historically higher healthcare needs raise concern that post-pandemic inequities in realised access to primary care may be widening, particularly for Pacific communities. Socioeconomic differences in respiratory presentation rates were also observed across the study period. Individuals living in areas of greater deprivation consistently demonstrated higher respiratory presentation rates, particularly for chronic lung disease. In contrast, milder acute conditions such as URTIs showed smaller socioeconomic gradients, potentially reflecting greater influence of health-seeking behaviour, employment-related certification requirements and other social factors on presentation patterns. Chronic respiratory disease presentations remained comparatively stable throughout the study period. Māori consistently experienced higher chronic lung disease presentation rates, likely reflecting the greater burden of smoking-related respiratory disease within Māori populations, where smoking prevalence remains substantially higher than the national average.29 Implications of the NLP methodology This study suggests the potential utility of applying NLP methods to routinely collected primary care EHR data to examine healthcare presentation patterns at the population scale. This approach addresses the international calls to focus on methods to analyse data from a health crisis to enable better responsiveness and appropriate public health measures and forward planning.30 The NLP approach used in this study incorporated both structured clinical data and free-text consultation records, enabling more comprehensive identification and classification of respiratory presentations across multiple diagnostic categories. This methodology enabled detailed examination of temporal respiratory presentation patterns across ethnic, socioeconomic and geographic groups throughout a rapidly evolving public health event. The rule-based NLP framework used in this study also provides transparency and interpretability in clinical classification processes, which may be advantageous for epidemiological surveillance and health services research. Although emerging machine learning and large language model approaches may further improve classification performance in future applications, the current methodology suggests that established NLP approaches can effectively support large-scale population analyses using routinely collected primary care data. Strengths A strength of this study is the ability to examine respiratory presentation patterns across a large primary care population over a 6-year period spanning the pre-pandemic, pandemic and post-pandemic phases of COVID-19. The dataset included more than 10 million general practice interactions across geographically diverse urban and rural practices, providing detailed longitudinal information on respiratory presentations within routine clinical care. The study also enabled examination of presentation patterns across multiple respiratory diagnostic categories and demographic groups using stable national measures of ethnicity and socioeconomic deprivation. The relatively stable overall volume of consultations across the study period strengthened comparisons between groups and over time. The use of validated NLP methods allowed analysis of both structured and unstructured clinical information that would not be accessible using conventional coded datasets alone. This approach provided detailed insight into respiratory presentation trends during a rapidly changing public health period and demonstrates the broader potential of routinely collected primary care data for population health monitoring and service planning. Limitations This study has several limitations. The data were derived from routinely collected general practice consultation records rather than a research-specific dataset and therefore reflect healthcare presentation rates (realised access) rather than true underlying disease incidence or unmet healthcare need. Changes in consultation modalities over time, including increased use of nurse-led and virtual consultations during the pandemic, may have influenced documentation practices and data capture. Although the NLP algorithm was validated, some degree of misclassification is inevitable, particularly for clinically overlapping respiratory conditions and evolving terminology associated with COVID-19-related presentations. Prioritised ethnicity classification may under-represent individuals with multiple ethnic affiliations, particularly those identifying as both Māori and Pacific. In addition, although participating practices represented multiple regions and included both urban and rural populations, the data set may not fully represent all NZ primary care settings. Finally, while the retrospective NLP approach enabled detailed longitudinal analysis, it was not designed for real-time surveillance or decision support within clinical systems. Conclusions This study identified differences in respiratory presentation patterns to NZ primary care according to ethnicity, socioeconomic deprivation and geographic region across the pre-pandemic, pandemic and post-pandemic periods. Pacific peoples, Māori and individuals living in areas of greater socioeconomic deprivation consistently experienced higher respiratory presentation rates, with the greatest disparities observed during periods of high SARS-CoV-2 transmission. NZ primary care services maintained relatively stable respiratory consultation volumes throughout the pandemic period and appeared responsive to increased healthcare needs among higher-risk populations during the acute phases of the pandemic. However, the decline in acute respiratory presentations observed post-pandemic, particularly among Pacific peoples, may indicate emerging barriers to primary care access and highlights the need for ongoing monitoring of healthcare inequities following major public health events. This study also highlights the potential value of applying NLP methods to routinely collected primary care EHR data to examine healthcare presentation patterns at the population scale. By incorporating both structured and unstructured clinical information, NLP approaches can provide detailed insights into temporal changes in healthcare utilisation and support population-level surveillance, service planning and health policy development. Leveraging routinely collected clinical data through validated analytical approaches may strengthen the ability of health systems to monitor evolving healthcare needs, identify inequities and support more equitable responses during future public health crises. Supplementary material