Exploring primary care providers’ perspectives on clinical decision support for venous thromboembolism diagnosis: a qualitative study

Abstract

Objectives

Diagnosing venous thromboembolism (VTE) in primary care is challenging due to non-specific symptoms and competing clinical demands. Although clinical decision support (CDS) tools are often proposed to improve diagnostic accuracy, their adoption remains inconsistent. This study aims to explore primary care providers’ (PCPs’) perspectives on diagnosing VTE and identify factors influencing the design, implementation and potential use of a VTE-specific CDS system in primary care.

Design

Qualitative interview study using directed content analysis.

Setting

Primary and urgent care settings across two academic health systems in urban and rural regions of the USA.

Participants

Eight PCPs, including physicians and physician assistants, were purposively sampled from two health systems. Participants practised in primary or urgent care settings and completed semistructured interviews focused on diagnosing VTE and CDS use.

Results

PCPs described diagnosing VTE as challenging because symptoms are often non-specific and encountered in time-constrained clinical environments. Participants viewed a VTE-specific CDS as potentially valuable, particularly if integrated seamlessly into the electronic health record and capable of supporting risk stratification using existing clinical data. However, clinicians also identified substantial concerns regarding alert fatigue, workflow burden, limited specificity and disruption of clinical autonomy. Medico-legal concerns are constructed as a prominent theme, with participants expressing apprehension that ignoring CDS recommendations could increase liability risk, while following alerts could contribute to unnecessary diagnostic testing and downstream consequences. Participants emphasised that CDS tools should be actionable, context-aware, minimally disruptive and supportive of rather than directive to clinical reasoning.

Conclusions

PCPs viewed diagnostic CDS for VTE as potentially beneficial but also potentially burdensome depending on how it is implemented within clinical workflows. Acceptance of CDS appears to depend not only on technical performance but also on its effects on workflow efficiency, clinician autonomy and perceived legal risk. Future CDS development for diagnostic decision-making in primary care should prioritise usability, workflow integration and minimisation of unintended consequences such as alert fatigue and over-testing.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This qualitative interview study explored primary care providers’ perspectives on venous thromboembolism (VTE) diagnostic clinical decision support (CDS) across two geographically distinct US health systems.

• The study focused specifically on diagnostic CDS in primary care, an area that has received less attention than therapeutic or guideline-based decision support systems.

• Use of a structured interview guide developed from prior work in VTE diagnostic delay and CDS research enhanced consistency across interviews.

• Directed content analysis combining deductive and inductive coding approaches enabled examination of predefined implementation domains while allowing the research team to construct additional themes through interpretation of participant narratives.

• The analysis was guided by predefined analytic domains, which may have reduced attention to concepts outside the initial coding framework.

Introduction

Primary care providers (PCPs) routinely make diagnostic decisions under conditions of uncertainty, time pressure and competing clinical demands.^1–4 These challenges are particularly pronounced when patients present with vague or non-specific symptoms, requiring clinicians to balance the risk of missing serious conditions against the risks of over-testing. Venous thromboembolism (VTE), including pulmonary embolism and deep vein thrombosis, exemplifies this diagnostic difficulty.⁵ Since patients with VTE often present with vague or non-specific symptoms, providers may not consider VTE in the differential diagnosis.

Although validated clinical tools such as the Wells criteria and Pulmonary Embolism Rule-out Criteria (PERC) exist,⁶ they are typically applied only when clinicians already suspect VTE. In practice, VTE may not be considered early in the diagnostic process, contributing to missed or delayed diagnosis. Our previous work showed that delayed and missed VTE diagnosis can increase all cause mortality.⁷ Efforts to improve diagnostic timeliness have included the development of electronic clinical quality measures and algorithms using electronic health record (EHR) data to identify patients at risk for delayed diagnosis.^7–10

Clinical decision support (CDS) systems offer a potential strategy to address these challenges by prompting consideration of VTE, supporting risk assessment and facilitating more targeted diagnostic testing.¹¹ While CDS technology is well established and increasingly integrated into EHR systems, its effectiveness depends on clinician acceptance and use in real-world settings. Prior research has largely focused on technical performance and guideline adherence, with less attention to how clinicians perceive diagnostic CDS in the context of everyday primary care practice.¹²

The aim of this study was to explore PCPs’ perspectives on VTE diagnosis and to identify factors that may influence the design and implementation of a VTE-specific CDS system in primary care.

Methods

Study design and participants

This study used a pragmatic qualitative descriptive approach and sought clinically relevant descriptions from PCPs who shared their experiences and perspectives concerning VTE diagnosis and VTE-focused CDS. This methodological orientation was selected to remain close to participants’ accounts while identifying practical considerations for CDS design and implementation. The research team employed semistructured interviews with PCPs from two health systems (Mass General Brigham (MGB) and the University of Alabama at Birmingham (UAB)), using a standardised interview guide developed by the team. The target population comprised clinicians providing primary or urgent care within the two participating health systems. Eligible participants were physicians or physician assistants who were actively practising in a primary or urgent care setting at MGB or UAB; no minimum level of clinical experience was required. Criterion-based purposive sampling was used to recruit clinicians whose professional role, practice setting and institutional affiliation provided experience relevant to the study aims. Potential participants were identified through institutional directories. Institutional permission was obtained to use the directories for research recruitment. Eligible clinicians were contacted by email and provided with a study information sheet before the interview. No reminder emails were sent. Recruitment continued until thematic sufficiency was reached, defined as the point at which no substantially new concepts relevant to the study aims were constructed from subsequent interviews.

Data collection

A structured interview guide was used to conduct interviews with PCPs. The interview guide was informed by the research team’s prior work in developing CDS systems^13–15 and electronic clinical quality measures for delayed VTE diagnosis (DOVE).^7–10 The final guide focused on PCPs’ experiences diagnosing VTE in primary and urgent care settings, perceived diagnostic challenges and facilitators, current use of decision-support tools and perspectives on the potential role, design and implementation of a VTE-specific CDS (online supplementary material 1).

Interviews were conducted remotely via secure videoconferencing platforms (Microsoft Teams and Zoom) by researchers experienced in clinical informatics and qualitative health services research. At the beginning of each interview, participants were given an opportunity to ask questions, and verbal informed consent was obtained before audio recording and data collection commenced. The interviewers had prior CDS and VTE research experience and had no supervisory, evaluative or other hierarchical relationships with participants. A structured interview guide ensured consistency while allowing open-ended responses. The recording/transcription functions of both MS Teams and Zoom were used. Thus, the interviews were automatically audio-recorded and transcribed verbatim. The transcripts were reviewed, and any information that could potentially identify either subject or site was obliterated or de-identified. Interview data were stored in the password-protected DOVE qualitative database and retrieved for analysis. Audio recordings and other study files were transferred to a secure institutional server immediately after collection. Transcripts and processed, de-identified files remained on the secure institutional server throughout data processing and analysis. A formal version-control system was not used.

A brief sociodemographic questionnaire administered employing the online survey tool (REDCAP) was used to collect participant age, gender, professional role, years of clinical experience, years at the current practice site, average weekly work hours and self-reported VTE knowledge and confidence in timely diagnosis (0–10 scale). Participants were compensated with a $50 gift card. Institutional review boards at both MGB (urban site) and UAB (urban/rural) approved the study protocol.

Data analysis

Interview texts were cleaned to remove any potentially identifying information, and providers were assigned pseudonyms (eg, participant 1). Interview transcripts were analysed using directed content analysis, a qualitative approach in which initial coding categories are derived from prior theory or empirical literature and subsequently applied to qualitative data while additional concepts are inductively constructed through coding and interpretation.¹⁶

Four a priori analytic domains were established based on prior CDS design and implementation research^{17 18}: (1) positive/desirable attributes, (2) negative/avoidable features, (3) perceived difficulty related to development, implementation or use of CDS and (4) considerations and recommendations for CDS design and implementation. These domains reflect constructs described in the CDS literature concerning usability, workflow integration, clinician acceptance and barriers to CDS adoption. Operational definitions of these domains and associated coding categories were developed prior to analysis (online supplementary material 2).

All transcripts were reviewed in full. To focus the analysis on discussions relevant to CDS development, transcript segments referencing CDS or related decision-support tools were initially identified. Because the interview guide covered multiple topics related to VTE diagnosis and quality measurement, this step allowed the research team to systematically identify text segments relevant to CDS while preserving analytic focus. When CDS-related segments were identified, adjacent transcript text was also reviewed to ensure that relevant contextual comments regarding diagnostic reasoning or workflow considerations were not excluded.

Two investigators (VKB and ACH) independently reviewed the extracted transcript segments. Analysis proceeded in two stages. First, segments were deductively assigned to one or more of the four broad a priori domains derived from previous CDS literature. Second, within these domains, the investigators constructed more specific concepts inductively through repeated comparison and interpretation of participants’ accounts. These concepts were not specified in advance and included workflow integration, alert fatigue, medico-legal concerns, diagnostic uncertainty, usability, accuracy and actionability. The inductive component was therefore bounded by the CDS-focused analytic scope and the four organising domains; an unrestricted inductive analysis of the complete interview dataset was not undertaken. Coding discrepancies were discussed and resolved through consensus, and transcripts and coded segments were reviewed repeatedly to support consistent interpretation.

Throughout analysis, the research team remained aware that their prior experience with CDS development and health informatics could influence interpretation of participants’ perspectives. Independent coding and regular discussions among investigators were used to promote reflexivity and ensure that interpretations remained grounded in participants’ statements.

A total of 36 transcript segments referencing CDS-related concepts were identified across the interviews, yielding 88 coded statements distributed across the four analytic domains. Individual segments sometimes contained multiple relevant concepts and therefore contributed to more than one coded comment. The coded data were subsequently examined to identify patterns across participants and to synthesise findings relevant to CDS design and implementation in primary care. The analysis focused on patterns across the full sample; findings were not formally stratified by professional role, practice setting or health system because the study was not designed for subgroup comparison.

The adequacy of the sample size was assessed using the principle of thematic sufficiency, defined as the point at which successive interviews yielded no substantially new concepts relevant to the study aims. Formal coding and directed content analysis were not conducted concurrently with data collection. During recruitment, clean verbatim transcripts were prepared, and the interviewers assessed whether each successive interview introduced new perspectives regarding VTE diagnosis or VTE-focused CDS. Later interviews largely reinforced previously identified concepts and did not yield substantively new perspectives. Based on these assessments, the research team determined that thematic sufficiency had been achieved after eight interviews and concluded recruitment. Information power was not applied prospectively as a formal stopping criterion but was considered as a complementary assessment of sample adequacy, taking into account the focused study aim, the relevance of participants’ clinical experience and the depth of the interview data. Formal directed content analysis was subsequently conducted using the complete set of transcripts. Previous research also supports the use of samples of this size for in-depth exploration of clinician perspectives.^{19 20}

To enhance analytic rigour, coding was performed independently by two reviewers with experience in CDS development and qualitative research. Regular discussions were conducted during the analytic process to review coding decisions and ensure consistent interpretation of the analytic domains. Reflexive consideration of investigator perspectives was maintained throughout the analysis to minimise the influence of prior assumptions regarding CDS development and implementation.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Results

Six participants were physicians, and two were physician assistants; four were male and four were female. Five participants identified as White, one as Asian and two as mixed race. The median age was 41 years (IQR 15.5). Participants had a median of 10 years of experience in their current profession (IQR 17) and 3 years at their current practice site (IQR 7). Median weekly work hours were 40 (IQR 0). On a scale from 0 to 10, the median self-reported VTE knowledge score and confidence in timely VTE diagnosis were both 8 (IQR 1). Participant characteristics are summarised in table 1.

Table 1. Characteristics of participants.

Characteristics	Value
Professional role
Physician	6 (75%)
Physician assistant	2 (25%)
Gender
Male	4 (50%)
Female	4 (50%)
Race
White	5 (63%)
Asian	1 (12%)
Mixed	2 (25%)
Age, years (median, IQR)	41 (15.5)
Experience in current profession, years (median, IQR)	10 (17)
Experience in current place, years (median, IQR)	3 (7)
Average hours worked, per week (median, IQR)	40 (0)
Self-reported VTE knowledge (0–10 scale), (median, IQR)	8 (1)
Self-reported confidence in timely diagnosis (0–10 scale), (median, IQR)	8 (1)

VTE, venous thromboembolism.

Across interviews, PCPs described VTE diagnosis as highly uncertain due to non-specific symptoms, time pressure and competing demands. While PCPs saw potential value in VTE-focused CDS, its acceptance depended on key tensions beyond technical performance, namely, workflow burden, clinical autonomy and medico-legal risk. These factors determined whether CDS was viewed as supportive or as added pressure in an already constrained environment. Together, these factors shaped whether CDS was perceived as a supportive aid or an additional source of pressure in an already constrained clinical environment. Example text segments for four categories are provided in online supplementary material 2).

PCPs’ were particularly apprehensive in the context of the current primary care crisis (shortage of PCPs and assistive staff, high turnover, high dissatisfaction, high burnout).²¹ PCPs fear litigation (participant 4 spoke of legal fears ‘But then you may get into the legalities of if you were given a reminder and you didn't act on it or you didn't feel it was relevant. Is that going to put you in a legal bind?’)—expressing concern if PCP judgement overruled a reminder that could be used in a malpractice lawsuit. PCPs were concerned about workflow burden (participant 5 spoke of not adding work ‘there’s a lot of concern…‘please don’t ask me to do one more thing’’) and alert fatigue (participant six stated ‘You know, we just signal the noise - alert fatigue’)—noting the risk of overrides and noise. Several PCPs emphasised that a CDS must be embedded into the EHR, seamless, actionable, trusted and context-aware, not just accurate. Participants suggested that a VTE-focused CDS should combine presenting symptoms, such as dyspnoea, chest pain, leg pain and unilateral leg swelling, with relevant risk factors, including recent surgery, prior VTE, malignancy, immobility or travel, hypercoagulable conditions and oestrogen use. They favoured a tool that automatically retrieves relevant information from the EHR and supports risk stratification, including the Wells criteria, rather than requiring clinicians to use external devices. Participants also emphasised that the CDS should connect risk assessment with appropriate diagnostic options, such as D-dimer testing, duplex ultrasonography or CT pulmonary angiography, while preserving clinician judgement and avoiding automatic testing pathways that could contribute to over-testing and diagnostic cascades. PCP voices revealed the tension between redundancy and imperative. Many PCPs acknowledged that CDS tools have existed for decades but see a VTE-specific CDS as potentially imperative because: VTE has non-specific, easily missed symptoms (participant 1, participant 2, participant 8). Tools including VTE diagnostic criteria, for example, Wells prediction rule and PERC,^{11 22} are not integrated into the EHR (participant 8 stated ‘when someone comes in right now I have a smart phone that has the criteria in it that I go through, but it’s not a tool that pulls anything from the chart’), contributing to wasted time and inefficiency in busy workflows. Missed VTE can have life-threatening consequences (participant 8 stated ‘for something that can be life threatening, I do think that a well created alert system could be helpful’). Yet, some PCPs view CDS as redundant when they already know the risk factors (participant 4 stated ‘I find those tools to be deceptively unhelpful because they’re only beneficial to the standpoint that you’re applying them’). Participant 6 questioned the possibility of extending a VTE CDS from aiding preliminary diagnosis to prevention via patient education (‘Do we need CDS to encourage people when they’re going on an aeroplane to tell them to walk around and drink more fluid? Now, that’s not diagnosis. That’s prevention’).

Discussion

Diagnosing VTE in primary care is challenging because symptoms are often non-specific and may overlap with more common conditions.⁹ In this qualitative study, PCPs described VTE diagnosis as a process shaped not only by clinical knowledge but also by competing cognitive, workflow and systemic pressures. Although CDS tools are often promoted as a means of improving diagnostic accuracy, our findings suggest that clinician acceptance depends on how these tools interact with the realities of primary care practice. Participants did not evaluate CDS solely based on accuracy or usability; instead, they considered how such tools might affect their clinical autonomy, workflow efficiency and exposure to medico-legal risk.

Participants emphasised that CDS tools are most useful when seamlessly integrated into the EHR and provide actionable support for decision-making. Consistent with prior research,^{23 24} tools embedded within existing workflows and requiring minimal additional effort are more likely to be adopted. Our findings further suggest that VTE-focused CDS should combine presenting symptoms with relevant clinical risk factors rather than rely on a single nonspecific symptom as an alert trigger. Automated retrieval of EHR data and support for established risk-stratification tools could reduce manual chart review and facilitate appropriate diagnostic testing. However, the symptoms, risk factors and testing pathways identified by participants should be regarded as clinician-informed design considerations rather than validated CDS criteria.

The workflow concerns identified in this study do not appear to be unique to VTE-focused CDS. International evidence across diagnostic, risk-assessment and disease-management applications in primary care has similarly identified increased workload, disruption of consultation flow, additional data entry and poor integration with existing clinical systems as barriers to CDS use.^{18 25} A scoping review of electronic CDS in general practice found that workflow effects depended substantially on how and when the tool was delivered, with interruptive or stand-alone systems more likely to add work or disrupt consultations than tools integrated into existing clinical processes.²⁵ Evidence from cancer diagnostic CDS has likewise identified workflow impact, trust in system recommendations and tension between decision support and clinicians’ responsibility for determining appropriate investigation or referral thresholds.²⁶ These findings suggest that clinicians’ preference for seamless, clinically relevant and actionable support represents a broader requirement for CDS across clinical areas rather than a preference unique to VTE. However, VTE-focused CDS may present additional challenges because nonspecific symptoms may initiate urgent testing and potentially lead to unnecessary diagnostic cascades.

Participants also identified barriers beyond usability. Alert fatigue and workflow burden were prominent concerns, reflecting limited tolerance for additional tasks. Consistent with prior studies,^{27 28} frequent or low-value alerts can lead to high override rates and reduced engagement with CDS. Participants also stressed the need to preserve clinical autonomy, noting that recommendations may feel redundant when risk factors are already recognised or intrusive when they appear to direct decision-making.

Our findings highlight an underexplored aspect of diagnostic CDS: its interaction with medico-legal concerns. Participants worried that ignored alerts could be scrutinised legally, while following them might trigger diagnostic cascades, incidental findings and low-value care. This dual risk fostered ambivalence and may encourage defensive decision-making. Although rarely addressed in CDS implementation research,²⁹ medico-legal considerations appear central to real-world adoption.

The findings also underscore the broader context of diagnostic uncertainty in primary care. Common, often benign symptoms (eg, dyspnoea or leg discomfort) may delay consideration of VTE.⁹ Participants suggested that CDS tools that surface relevant risk factors or prompt consideration of VTE in appropriate contexts could support clinical reasoning, especially when presentations are vague or atypical.

This study extends CDS research by focusing on diagnostic decision support in primary care rather than therapeutic or guideline-based tools. While prior work has emphasised improving treatment adherence,^30–32 our findings highlight challenges unique to supporting diagnosis of conditions with non-specific presentations. The interplay of workflow constraints, clinical autonomy and medico-legal concerns suggests that successful CDS implementation must address more than technical accuracy. This work also builds on research into DOVE, including our DOVE measure, by exploring how CDS might support earlier recognition in primary care.^33–35 Overall, these findings inform the design and implementation of diagnostic CDS aimed at reducing delays.

Future work may explore how emerging AI methods could support CDS tools designed to assist diagnostic decision-making. AI-enabled systems may enhance the ability to integrate structured and unstructured EHR data, identify subtle risk patterns and present context-specific recommendations. However, the concerns identified in this study suggest that technological advances alone will be insufficient unless they are aligned with clinician workflows and address broader issues related to usability, trust and perceived risk.

Limitations

Several limitations should be considered. The study included a small number of participants, and therefore, the findings should not be interpreted as representing the perspectives of all PCPs. However, participants were recruited from two distinct health systems, allowing perspectives to be captured across different clinical environments. Recruitment relied on voluntary participation which may introduce selection bias as clinicians with particular interest in diagnostic decision-making or CDS tools may have been more likely to participate. Although participants were recruited from two health systems and included both physicians and physician assistants, the sample was not designed or powered to examine differences by health system, practice setting or professional role. The heterogeneity of the sample may also have reduced the information power available for subgroup-specific interpretation. In addition, the analysis focused on transcript segments related to CDS and was guided by predefined analytic domains which may have influenced the identification and interpretation of themes.

Conclusion

PCPs in this study viewed diagnostic CDS as both potentially helpful and potentially burdensome, depending on how it interacts with clinical workflow, decision-making and perceived legal risk. These findings suggest that successful implementation of diagnostic CDS in primary care will require more than technical accuracy. Tools must be designed to support clinician judgement, minimise unnecessary alerts and address concerns related to over-testing and medico-legal exposure. Recognising these broader influences may be essential to improving adoption and realising the potential of CDS to enhance diagnostic safety.

Data availability statement

No data are available.