Validation of a standardised approach to collect sociodemographic and social needs data in Canadian primary care: cross-sectional study of the SPARK tool

Abstract

Objective

This study validates the previously tested Screening for Poverty And Related social determinants to improve Knowledge of and access to resources (‘SPARK Tool’) against comparison questions from well-established national surveys (Post Survey Questionnaire (PSQ)) to inform the development of a standardised tool to collect patients’ demographic and social needs data in healthcare.

Design

Cross-sectional study.

Setting

Pan-Canadian study of participants from four Canadian provinces (SK, MB, ON and NL).

Participants

192 participants were interviewed concurrently, completing both the SPARK tool and PSQ survey.

Main outcomes

Survey topics included demographics: language, immigration, race, disability, sex, gender identity, sexual orientation; and social needs: education, income, medication access, transportation, housing, social support and employment status. Concurrent validity was performed to assess agreement and correlation between SPARK and comparison questions at an individual level as well as within domain clusters. We report on Cohen’s kappa measure of inter-rater reliability, Pearson correlation coefficient and Cramer’s V to assess overall capture of needs in the SPARK and PSQ as well as within each domain. Agreement between the surveys was described using correct (true positive and true negative) and incorrect (false positive and false negative) classification.

Results

There was a moderate correlation between SPARK and PSQ (0.44, p<0.0001). SPARK correctly classified 71.4% of participants with or without a social need. There was strong agreement with most demographic questions. SPARK correctly classified 74.3% of participants as having financial insecurity. Clustering financial security questions had fewer false negatives (6.4%, n=11/171 vs 9.9%, n=17/171) and more false positives (19.3%, n=33/171 vs 11.1%, n=19/171) when compared with the question ‘difficulty making ends meet’. When looking specifically at participants with high UCLA loneliness scores (>60), SPARK correctly classified 90.5% (n=176/191).

Conclusions

SPARK provides a brief 15 min screening tool for primary care clinics to capture social and access needs. SPARK was able to correctly classify most participants within each domain. Related ongoing research is needed to further validate SPARK in a large representative sample and explore primary care implementation strategies to support integration.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Screening for Poverty And Related social determinants to improve Knowledge of and access to resources (SPARK) is a previously tested tool that can support collection of demographic (language, immigration, race, indigenous identity, sex, gender identity, sexual orientation and disability) and social (education, perceived financial strain, housing stability, medication adherence, transportation, social supports and precarious employment) information from patients.

• Purposive sampling was used to capture a diverse sample of adult participants from four Canadian provinces (Saskatchewan, Manitoba, Ontario and Newfoundland and Labrador) that varied by age, race, location, gender identity and experiences with social needs.

• We assessed concurrent validity of SPARK administered simultaneously with reference standard questions on the Post Survey Questionnaire to assess the ability of SPARK to estimate an individual’s current situation.

• Reference standard questions are defined as the best available method for capturing a condition; poor correlation with a reference standard may indicate that the reference standard does not completely or accurately estimate the true value.

• This study represents a relatively small sample (n=192) of English-speaking participants. Findings of this study may not represent the general Canadian population and may not be generalisable to other countries.

Introduction

Social determinants of health (SDH) are non-medical factors that influence an individuals’ well-being, health and health outcomes.^{1 2} The SDH can have both positive and negative effects on health equity; the opportunity to achieve full health potential.³ Systematic conditions and factors create unfair and avoidable differences in health status, with health conditions more prominent in lowest socioeconomic groups.¹,⁶

In Canada, healthcare delivery is a provincial/territorial responsibility with medically necessary services provided by single-payer, publicly funded insurance. To receive federal funding for health services, provinces and territories must meet standards described in the Canada Health Act: public administration, comprehensiveness, universality, portability and accessibility.⁷ In Canada, primary care represents an individual’s first point of contact with the healthcare system. It includes acute care as well as primary and secondary prevention, chronic disease management and health maintenance.^{8 9} An integrated primary care system as envisioned by Fuller requires streamlined access to proactive, personalised care from a multidisciplinary team.¹⁰ Similarly, team-based primary care continues to gain support with models including Patient Centered Medical Home (USA) and Patient Medical Home (Canada) where nurses and other providers complement care provided by physicians and nurse practitioners.^{8 9} Integrated primary care not only enables a more holistic approach to health and well-being but is pivotal to tackling health inequalities.^{10 11}

A crucial barrier to understanding and addressing health inequities is the lack of SDH data at an individual level, with the majority of research assessing aggregate gradients.⁴ Andermann found patients felt primary care providers were unaware of their social context including struggles with food security, transportation and medication access.³ Although clinical practice tools and training exist,³¹²,¹⁸ they are often focused on a specific population (eg, Indigenous persons, immigrants and refugees) or a specific determinant of health (eg, poverty, food insecurity, housing instability, mental health problems) and there is a lack of guidance on universal screening of asymptomatic patients.^{312 17 19},²³ Supporting continued improvements in primary care requires effective data collection and use within provider–patient interactions to inform personalised care, as well as clinic and system level actions.³,^{510 11 13} Despite the role of primary care in health system access, collection of social history information and the SDH is inconsistent and not standardised.^{3 4} The objective of this study is to assess and compare a brief demographic and social needs survey, titled Screening for Poverty And Related social determinants to improve Knowledge of and access to resources (SPARK) Tool to a collection of reference standard questions to establish a brief standardised tool for the collection of SDH in primary care settings.

Methods

SPARK was developed through multiple processes including research, consultations and collaboration with relevant organisations and groups.^{18 24} Between 2012 and 2019, SPARK was piloted, revised and tested.²⁰,²³ Following revisions, the current version of SPARK consists of 16 questions aimed at capturing a patient’s demographic and social needs at a primary care appointment (online supplemental appendix A).

Recruitment

Purposive sampling was used to capture a sample of adult participants from four Canadian provinces (SK, MB, ON and NL) that varied by age, race, location, gender identity and experiences with social needs.²⁴ Sample size estimates were calculated using PASS 2019 Power Analysis and Sample Size Software²⁵ assuming prevalence of 50% with two-sided 95% CI, a width of 16.5% and sample specificity of 80% suggests a sample size of 200 participants. Advertisements were posted online, emailed and displayed at primary care clinics and community centres. The study was shared by research team members within their local networks. Advertisements were available in English, French and translated into the three most commonly spoken languages in each province.^{24 26 27}

Data collection

Semistructured interviews (~60 min) were conducted with participants. Participants were asked to begin the interview by completing SPARK. Cognitive interviewing prompted participants to think-aloud while completing SPARK to provide feedback on the questions and how they selected their responses as described by Adekoya et al.²⁴ Immediately following completion of SPARK, participants were emailed a survey link and asked to independently complete the Post Survey Questionnaire (PSQ), a collection of seven distinct surveys used to assess different domains captured within SPARK.²⁶,³⁶ Interviewers remained available to the participant while they completed the PSQ in case they had any questions or difficulty submitting their responses.

Patient and public involvement

Patient and community partners were involved in all stages of this study. Patient and community partners are members of the research team and attend monthly team meetings. They are encouraged to share experiences and inform the research question, process and outcomes. Patient and community members assisted with the distribution of study invitations and advertisement. These partners reviewed and further revised study materials. For example, they participated in a mock interview to test the survey, interview questions and research process, suggesting adjustments incorporated by research coordinators. Patient and community partners reviewed the preliminary code book and assisted with interpretation of the results. They are coauthors on all publications and presentations of this work.

Tool components

SPARK domains

SPARK is composed of 16 questions within 10 domains (online supplemental appendix A). Demographic questions assess language, immigration status, race, disability, sex at birth, gender identity and sexual orientation. Social needs are captured within the following five domains: education, financial security, housing, social support and precarious employment. There are five questions capturing financial security: (1) making ends meet, (2) medication access, (3) transportation cost, (4) utility bill payments and (5) ability to pay mortgage or rent. Two questions assess social support; family and friends (1) you can open up to, and (2) rely on. Last, three questions assess precarious employment: (1) casual, short-term or temporary position, (2) fear you could be fired if you raise employment concerns and (3) pay that varies from month to month.

We identified two types of needs on SPARK. Access needs were represented by preference for translation (language), being a newcomer (immigration status) or having disability-related access needs. Social needs were represented by having less than high school education, financial insecurity, housing instability, no social support or precarious employment (online supplemental table 1).

PSQ domains

The PSQ includes 7 separate surveys with 72 questions representing the 10 domains captured by the SPARK tool (online supplemental appendix B, online supplemental table 1). The PSQ used questions from the 2016 Canadian Census^{26 27} and 2014 Canadian Community Health Survey²⁹ to assess language, immigration status, Indigenous identity, race and education. In the PSQ, disability was assessed using simple scoring of the 12-item WHO Disability Assessment Scale (WHODAS 2.0),²⁸ to assign a score (0–5) and categorise them into level of disability: none, mild, moderate, severe and extreme. Income was assessed using the low income cut-off (LICO), capturing cost of living with adjustment for seven family sizes and five community sizes.³⁰ Housing was assessed using the Veterans Affairs Homelessness Screening Clinical Reminder.³⁵ We used the 20-item University of California, Los Angeles (UCLA) Loneliness Scale³¹ to measure one’s subjective feelings of loneliness and social isolation with scores indicating: none 1–19, mild 20–40, moderate 41–60 or high >60 loneliness. The Poverty and Employment Precarity in Southern Ontario (PEPSO)³² employment precarity index assessed stability of employment with the 10-question survey providing a score between 0 and 100 (ie, secure employment <2.5, stable employment 3–17.5, vulnerable 18–37.5 and precarious employment ≥38).³² Sex at birth, gender identity and sexual orientation questions were informed by literature.^{33 34}

Statistical analyses

We assessed concurrent validity¹⁹ of SPARK administered simultaneously with the PSQ to assess the ability of SPARK to identify an individual’s current social and access needs. Exploratory factor analysis was used to assess our secondary outcome to explore the validity of different domains within SPARK as compared with their corresponding PSQ survey.

We compared SPARK to the PSQ by describing participant responses using frequency (%) and mean (SD). Bivariate correlation coefficients compared respondents with needs identified in SPARK to those with needs identified on the PSQ. We assessed the summary scores and sample characteristics using both dichotomous (yes/no) and polytomous (multilevel) scales using t-tests and χ² comparisons. The Pearson correlation coefficient (r) measures linear correlation between two sets of data. The Cohen’s kappa measure of inter-rater reliability (κ) assessed the agreement between dichotomised and categorical variables within each domain as well as correlation between overall measures explicitly adjusting for agreements that could be attributed to chance. Cramer’s V (φ_c) assessed the effect size measurement for two categorical variables with multiple categories. The kappa statistic ranges from 0 (no agreement) to 1 (perfect agreement); 0.1–0.2 (slight agreement), 0.21–0.4 (fair agreement), 0.41–0.6 (moderate agreement), 0.61–0.8 (substantial agreement) and 0.81–0.99 (near perfect agreement). Significance was reported at alpha of 0.05. A coefficient of <0.29 has low correlation, 0.3–0.49 has moderate correlation and 0.5–1 has high correlation.

Diagnostic validity was assessed through agreement using contingency tables. Among categorical variables, we assessed agreement in the most prominent categorical group. True positive (TP) represents ‘yes’ on both SPARK and PSQ. True negative (TN) represents ‘no’ on both SPARK and PSQ. Correct classification indicates combined TP and TN. False positive (FP) responses were ‘no’ on SPARK and ‘yes’ on PSQ (type I error). False negative (FN) responses were ‘yes’ on SPARK and ‘no’ on PSQ (type II error). Incorrect classification indicates combined FP and FN. Sensitivity defines how well SPARK can identify TP cases. Specificity represents how well SPARK can identify TN responses. CIs for sensitivity, specificity and accuracy are ‘exact’ Clopper-Pearson CIs.

Statistical analyses were conducted using SAS V.9.4 (SAS Institute). Response categories of less than 5 were suppressed. Missing responses, including ‘not applicable’ were removed from domain-specific agreement and correlation analyses. Researchers used cognitive interviewing to administer the survey collecting data on responses provided and reasons for not responding. Descriptive, bivariate and thematic grouping²⁴ were reviewed to ensure data was missing at random. Cognitive interviewing during capture of the data allowed the researchers to validate data entry. Histograms and scatterplots were used to visualise the data and identify any residuals.

Results

There were 192 participants, 123 participants from Ontario, 24 from Saskatchewan, 24 from Manitoba and 21 from Newfoundland and Labrador. There was a greater number of participants in Ontario due to a larger population of ~15.8 million compared with ~1.4 million in Manitoba, ~1.2 million in Saskatchewan and ~540 000 in Newfoundland and Labrador.³³ On average, it took participants 15 min (SD 11) to complete SPARK. The majority of respondents were born in Canada (n=118/189, 62.4%), were women (n=125/192, 65.1%), indicated white race (n=53/167, 31.7%) and had a college or university degree (n=127/192, 66.1%) (table 1).

Table 1. Characteristics of participants.

SPARK tool question	Participants
Demographics
Race,* n (%)
White	53/167 (31.7)
South Asian or Indo-Caribbean	36/167 (21.6)
Black	23/167 (13.8)
East Asian	18/167 (10.8)
Arab, Middle Eastern or West Asian	16/167 (9.6)
Southeast Asian	14/167 (8.4)
Indigenous	10/167 (6.0)
Latino/Latina/Latinx	5/167 (3.0)
Another race	9/167 (5.4)
Sex at birth, n (%)†
Female	132/189 (69.1)
Male	57/189 (29.8)
Intersex	<5
Gender identity, n (%)
Women	125/192 (65.1)
Man	57/192 (29.7)
Transgender	<5
Gender fluid or gender non-binary	8/192 (4.2)
Sexual orientation, n (%)
Heterosexual	163/188 (86.7)
Gay	<5
Lesbian	<5
Bisexual	9 (4.8)
Queer or pansexual	10 (5.3)
Access needs
Born in Canada, n (%)	118/189 (62.4)
Translation would be helpful, n (%)	14/190 (7.4)
Disability-related access need (≥1 vs 0), n (%)	87/190 (45.8)
Difficulty seeing	34/190 (17.9)
Difficulty hearing	16/190 (8.4)
Difficulty walking or climbing	19/190 (10.0)
Difficulty remembering or with concentration	47/190 (24.7)
Difficulty with self-care	33/190 (17.4)
Difficulty communicating	30/190 (15.8)
Social needs
Education, n (%)
Completed grad school	<5
Some high school	<5
Completed high school or equivalent	8/192 (4.2)
Trades certificate/diploma	14/192 (7.3)
Some college/university	39/192 (20.3)
College or university degree	93/192 (48.4)
Postgraduate degree	34/192 (17.7)
Financial Insecurity,‡ n (%)	65/192 (33.9)
Difficulty making ends meet	43/192 (22.4)
Avoided filling a prescription or did something to make a prescription last longer	29/143 (20.3)
Difficulty getting to appointment because of the cost	13/170 (7.6)
Miss a utilities payment	17/160 (10.6)
Difficulty paying mortgage or rent, n, (%)§	13/144 (9.0)
Housing, n (%)
Own home	72/192 (37.5)
Rent	72/192 (37.5)
Staying with friends or relatives because no alternative	10/192 (5.2)
Shelter	<5
On the street	<5
Other	37/192 (19.4)
Social housing, subsidised housing or rent-geared-to-income, n (%)¶	11/72 (15.3)
Unstable housing, n (%)	22/192 (11.5)
Social support, n (%)**
Family or close friends you can open up to	177/192 (92.2)
Can rely on family or friends if need help	169/192 (88.0)
No social support	14/192 (7.29)
Precarious employment (≥2 yes), n (%)	38/180 (21.1)
Employed in a casual, short-term or temporary position, n (%)	46/173 (26.6)
Fearful of repercussions if raise employment concerns, n (%)	38/150 (24.7)
Pay varies a lot from month to month, n (%)	36/155 (22.6)
Needs identified
Social needs, mean (SD)	0.74 (0.9)
Social needs identified (≥1), n (%)	97 (50.5)
Social and access needs, mean (SD)	0.7 (0.9)
Social and access needs identified (≥1), n (%)	138 (71.9)

^*Race was not asked in one province, 171 participants were asked about race.

^†Intersex was represented by <5 participants and, therefore, was not included in the total.

^‡Not applicable responses were not included.

^§Difficulty paying mortgage or rent was only asked of people that indicated they owned a home or paid rent.

^¶Social housing, subsidised housing or rent-geared-to-income question was only asked of participants who paid rent.

^**Reflects participant responding to housing question indicating they are staying with friends or family, living in a shelter or on the street or indicated they are on social housing, subsidised housing or rent-geared-to-income.

SPARK, Screening for Poverty And Related social determinants to improve Knowledge of and access to resources.

Based on SPARK, 71.9% (n=138/192) of participants identified ≥1 need, with 50.5% (n=97/192) reporting ≥1 social need (table 1). On the PSQ, 49.0% (n=94/192) of participants indicated ≥1 need, with 41.7% (n=80/192) reporting a social need (online supplemental table 2). Association between SPARK and PSQ was stronger for social needs (κ=0.43) than social and access needs combined (κ=0.3) (table 2). SPARK correctly classified 71.4% (n=137/192) of participants with or without a social need, and 64.6% of participants with or without a social or access need (n=124/192). When identifying both social and access needs compared with social needs alone, the sensitivity increased (87.2% (95% CI 78.8% to 93.2%) vs 76.3% (95% CI 65.4% to 85.1%)) but the specificity decreased (42.9% (95% CI 32.9% to 53.3%) vs 67.9% (95% CI 58.4% to 76.4%) due to an increase in FP (18.8%, n=36/192 vs 29.2%, n=56/192) (online supplemental table 3). Online supplemental table 1 outlines the process for identifying social and access needs using the SPARK.

Table 2. SPARK compared with Post Survey Questionnaire (PSQ) correlation coefficient.

SPARK vs PSQ	Correlation coefficient
Needs identified
Social need (per 1 need increase)	r=0.568 (<0.0001)
Social needs identified (≥1 SPARK vs ≥1 PSQ)	κ=0.428
Social and access need (per 1 need increase)	r=0.576 (<0.0001)
Social and access needs identified (≥1 SPARK vs ≥1 PSQ)	κ=0.298
Demographic variables
Race (SPARK vs reference)	φc=0.882 (<0.0001)
Sex at birth (SPARK vs reference)	φc=0.994 (<0.0001)
Gender identity (SPARK vs reference)	φc=0.967 (<0.0001)
Sexual orientation (SPARK vs Reference)	φc=0.986 (<0.0001)
Access variables
Translation vs language other than English spoken most often at home	κ=0.16
Immigration status: Arrived ≤5 years ago (SPARK vs reference)	κ=0.634
Disability-related access need (≥1) vs Disability (≥1)	κ=0.359
Disability-related access need (≥1) vs Disability score ≥3	κ=0.477
Social needs: education
Less than high school (SPARK vs reference)	κ=0.604
Social needs: income
Financial insecurity vs income <US$30 000	κ=0.368
Financial insecurity vs below low-income cut-off	κ=0.332
Difficulty making ends meet vs income <US$30 000	κ=0.379
Avoided filling a prescription or did something to make a prescription last longer vs income <US$30 000	κ=0.356
Difficulty getting to appointment because of the cost vs income <US$30 000	κ=0.236
Miss a utilities payment vs income <US$30 000	κ=0.351
Difficulty paying mortgage or rent vs income <US$30 000	κ=0.236
Social needs: unstable housing
Unstable housing (SPARK vs reference)	κ=0.319
Unstable housing vs unstable housing and worried or concerned about stable housing	κ=0.323
Social needs: social support
Social isolation (SPARK (≥1) vs high social isolation score >60)	κ=0.11
Social isolation (SPARK (≥2) vs high social isolation score >60)	κ=0.179
Social isolation (SPARK (≥2) vs moderate or high social isolation score >41)	κ=0.07
Precarious employment
Precarious employment (SPARK (≥1) vs PEPSO≥38)	κ=0.436
Precarious employment (SPARK (≥2) vs PEPSO≥38)	κ=0.405

PEPSO, Poverty and Employment Precarity in Southern Ontario; r, Pearson correlation coefficient; SPARK, Screening for Poverty And Related social determinants to improve Knowledge of and access to resources; κ, kappa statistic; φ_c, Cramer’s V.

Demographic and access variables

Most demographic questions had strong agreement between SPARK and PSQ (figure 1, online supplemental table 3). Table 2 demonstrates the high correlation between the two surveys on immigration, race, sex at birth, gender identity and sexual orientation, with SPARK able to correctly classify 88.7%–99.5% of individuals in these areas.

Figure 1. SPARK demographic and access need responses compared with the Post Survey Questionnaire. SPARK, Screening for Poverty And Related social determinants to improve Knowledge of and access to resources.

Participants’ response to translation services was mixed, with 22.3% speaking a language other than English at home (PSQ) but only 16.7% reporting needing translation services (SPARK). However, SPARK was able to correctly classify 78.0% of participants as needing or not needing translation (figure 1, online supplemental table 3).

Almost half (45.8%, n=87/190) of respondents indicated that they had a disability-related access need on SPARK (table 1), with moderate agreement between the PSQ and SPARK (κ=0.36) (table 2). SPARK correctly classified 66.7% (n=126/189) of the respondents as having a difficulty (figure 1), with a moderate sensitivity (58.4% (95% CI 49.7% to 66.8%)) due to high FN (30.2% (57/189)) (online supplemental table 3). Correlation slightly improved among participants who reported a disability score ≥3 on the PSQ (κ=0.48) with 70.1% (n=140/189) of participants classified correctly (table 2).

Social needs

Using SPARK, half of the participants (50.5% (n=97/192)) indicated ≥1 social need; 32.8% had one social need and 5.7% had ≥2 social needs. More specifically, 22.4% (n=43/192) indicated they had difficulty making ends meet, 9.0% (n=13/144) indicated they had difficulty paying their mortgage or rent and 21.1% (n=38/180) indicated precarious employment (table 1).

Financial security questions: Clustering financial insecurity questions (≥1 positive response) had low to moderate agreement with <US$30 000 (κ=0.37) and LICO (κ=0.33) (table 2). The SPARK ‘difficulty making ends meet’ question had the highest correlation with income (κ=0.38), followed by ‘avoided filling a prescription’ (κ=0.36) and ‘missing a utilities payment’ question (κ=0.35) (table 2). When compared with the ‘difficulty making ends meet’ question, clustering financial security questions had fewer FN (6.4%, n=11/171 vs 9.9%, n=17/171) and more FP (19.3%, n=33/171 vs 11.1%, n=19/171) (figure 2, online supplemental table 3). Notably, 10.4% (n=20) of participants responded to ≥1 SPARK financial question but indicated ‘prefer not to answer’ on the PSQ income question. These participants were not included in correlation estimates. Overall, SPARK correctly classified 74.3% (n=127/171) of participants as financially secure or insecure compared with <US$30 000 and 71.9% (n=123/171) compared with LICO (figure 2, online supplemental table 3). Identification of low-income (PSQ) using SPARK required ≥1 positive response to a financial security question. The financial security questions on SPARK provide additional information regarding access to care in addition to income.

Figure 2. SPARK social need responses compared with the Post Survey Questionnaire. SPARK, Screening for Poverty And Related social determinants to improve Knowledge of and access to resources.

Housing stability: In the PSQ, unstable housing was captured with two questions. SPARK correctly classified 85.7% of participants with ≥1 positive response (κ=0.32, p<0.0001) and 90.5% of participants with 2 positive responses (κ=0.32) by asking about current housing (table 2). When we assessed the agreement, sensitivity was significantly lower with two PSQ questions (39.1% (19.7%–61.5%) vs 83.3% (35.9%–99.6%)) associated with an increase in FN (7.4%, n=14/189 vs 0.5%, n=1/189) (online supplemental table 3).

Social support: The correlation between SPARK and UCLA loneliness score was stronger when 2 positive responses were required on SPARK (κ=0.18) compared with ≥1 positive response (κ=0.11) (table 2). When looking specifically at participants with high UCLA loneliness scores (>60), SPARK correctly classified 90.5% (n=176/191) when both SPARK questions were used, compared with 85.7% with ≥1 SPARK question (figure 2, online supplemental table 3). Participants with moderate (41–60) or high (>60) social isolation scores were not well captured by SPARK, with 50.8% correctly classified due to high FN (48.6%, n=91/191) (online supplemental table 3).

Precarious employment: There was a moderate correlation between SPARK precarious employment questions and having a high PEPSO score (r=0.47, p<0.0001), with 80.0% of participants (n=148/185) correctly classified (figure 2, table 2). When ≥1 positive response was required on SPARK, we saw high sensitivity (85.4% (70.8%–94.4%)) and moderate specificity (71.5% (63.4%–78.7%)). However, when two positive responses were required, the sensitivity decreased to 51.2% (35.1%–67.1%) due to an increase in FN (10.8%, 20/185 vs 3.2%, 6/185) but specificity increased (88.2% (81.7%–93.0%)) associated with a decrease in FP (9.2%, 17/185 vs 22.2%, 41/185). In practice, capturing patients with ≥2 positive responses will prevent overcapture of patients that do not have precarious employment (online supplemental table 3).

Discussion

SPARK is a brief 15 min screening tool for primary care clinics to capture SDH and identify social and access needs, collecting reliable information within one questionnaire to inform personalised care. Overall, there was a moderate correlation between SPARK and PSQ in identifying patients with social or access needs. On exploratory factor analysis, we found high agreement among most demographic questions, with greater variation in agreement among social needs questions. SPARK was able to correctly classify most participants within each domain and particularly patients with higher levels of disability or social isolation. Within primary care clinics, SPARK provides an opportunity to identify a variety of different access and social needs using one screening tool.

SPARK aims to identify difficulties and barriers accessing health services that can be addressed at primary care appointments. Therefore, questions are specific to the context of care provision.²² Contrary to Federici et al,³⁷ the moderate agreement between SPARK and WHODAS 2.0 may demonstrate a difference in survey intent. Pinto et al found discrepancies between patient self-reported and provider-documented disabilities, suggesting patients misjudge what constitutes a disability and may be concerned about stigma.²³ Disability questions and response options should be clearly defined to reduce misinterpretation and ensure accurate capture.^{23 24} Despite this, SPARK was able to capture over 70% of participants with more severe disability according to WHODAS 2.0.

This study found differences in capture of social needs with low-to-moderate correlation between SPARK and PSQ. SPARK captures perceived financial strain or financial worries as opposed to self-reported income and LICO collected on the PSQ. Perceived financial strain captures fluctuating financial circumstances that are associated with physical and psychological well-being, academic and social integration, self-rated health, depression and functional impairment. Financial strain is more pronounced among unmarried, unemployed, lower-income households and renters but has low-to-moderate correlation with objective measures of income.^{20 38} Similar to this study, Pinto et al found higher response rates to questions about financial strain when compared with objective measures of income.²⁰

Implications for research and practice

Primary care teams and patients are supportive and willing to incorporate brief screening surveys in primary care settings.314 18 21 22 39,⁴¹ Further, interactive communication technology for primary care has the potential to improve preventative services through screening, prioritising and initiating interventions.^{15 17} To realise this potential, the tools delivered through the technology need to be valid. However, a systematic review by Sokol et al found that the majority of studies screening for SDH did not assess the psychometric properties and validity of the screening tool, instead focusing on immediate outcomes from the screen as well as patient and provider perspectives.⁴² SPARK addresses topics prevalent on other screening tools including employment, financial security, medication access and housing status which influence health outcomes and identify populations that may benefit from targeted interventions.^{12 14 43 44} Further research should explore barriers to incorporating SPARK into workflow, including perceptions, communication challenges, resource allocation and capacity to address SDH.³³⁹,⁴²

Limitations

This study assessed concurrent validity of SPARK to reference standard questions on the PSQ; therefore, results of this study cannot assess potential or future performance. Reference standard questions are defined as the best available method for capturing a condition compared with gold standard questions which are established and validated method for capturing a condition. Poor correlation with a reference standard may indicate that the reference standard question does not completely or accurately estimate the true value. Although we did not capture the time to complete the PSQ in the study, it is estimated as taking 30 min to complete. All participants completed the SPARK prior to the PSQ, which may have contributed to participant fatigue and response bias when completing the PSQ. Findings of this study may not represent the general Canadian population and may not be generalisable to other countries. Despite offering translation to participants, all surveys and interviews were conducted in English, which may have biased study results, especially the validity of the translation question. In addition, although the research team aimed to capture participants from different educational backgrounds and racial identities, the majority of participants had postsecondary education and were white. Lastly, the relatively small sample size, particularly among some of our subgroups, prevented subgroup analyses. Further research should validate SPARK on a larger, representative sample.

Conclusions

There is an increasing recognition that to address health inequity and improve population health, SDH need to be systemically captured and prioritised within healthcare settings.³ Organisational readiness and system-level mandates may accelerate adoption and implementation of SDH data collection using a validated tool.^{9 41} Further refinement of SPARK based on these findings as well as additional research exploring its implementation and outcomes will improve capture of SDH in primary care and support wider implementation in clinical practice. Collection and use of SDH promotes health equity and ensures sustainability of a publicly funded healthcare system.⁴

Data availability statement

Data are available on reasonable request.