Elinor Nemlander

Background: Diagnostic safety concerns are among the most consequential yet least visible threats to patient safety in primary care. Non-specific symptoms potentially associated with cancer represent a high-risk context, where delayed diagnosis may have serious consequences despite a low probability of serious disease. Objective: To identify diagnostic safety concerns and areas for improvement in assessing patients presenting with non-specific symptoms in primary care. Method: A structured medical record review combining quantitative assessment and qualitative thematic analysis was conducted in Region Stockholm, Sweden. In a region-wide quality improvement initiative, 27 primary care centres reviewed 1116 medical records of patients aged 40–75 years with selected ICD-10 (International Classification of Diseases, 10th Revision) codes recorded in 2023. Reviews used a structured template addressing diagnostic reasoning, data collection and healthcare use. Quantitative data were analysed descriptively, and qualitative data were analysed using reflexive thematic analysis. Results: Overall, 358 records (32%) had at least one diagnostic safety concern. Absence of documentation of differential diagnostic reasoning was identified in 21% of records, insufficient data collection in 18% and failure to consider alternative diagnoses in 12%; these concerns co-occurred in 13% of records. Four themes were identified: invisible diagnostic reasoning, narrow diagnostic framing, failure to re-evaluate, and fragmented care and patient complexity as a cross-cutting contextual influence on diagnostic vulnerabilities. Conclusion: Among patients presenting with non-specific symptoms potentially associated with cancer, documentation-related diagnostic safety concerns arose from whether and how diagnostic reasoning was documented, revisited and communicated across encounters. While the findings support safety-netting and diagnostic re-evaluation, they also highlight vulnerabilities including negative test results functioning as diagnostic endpoints and care fragmentation undermining diagnostic progression. Structured record review linked to reflective learning may help address these patterns in practice.

Background: As organised colorectal cancer (CRC) screening programmes expand across Europe, little is known about their downstream effects on healthcare utilisation beyond the screening pathway. This study aimed to compare changes in healthcare utilisation among individuals with positive screening results without diagnosed CRC and participants with negative screening results..Methods: We conducted a longitudinal cohort study within a Swedish population-based CRC screening programme using faecal occult blood tests (FOBT), among individuals aged 60–69 years. Individual-level registry data linked across multiple national and regional databases were analysed using a difference-in-differences design with Poisson regression. Screening participants between 2008 and 2021 were included, whereas individuals with diagnosed CRC were excluded. Four outcomes were pre-specified: outpatient visits in primary and specialist care for gastrointestinal diseases or psychiatric health disorders, all-cause primary care visits, gastroscopy procedures, and dispensed prescription drugs.Findings: The sample consisted of 13,400 individuals with positive FOBT results and 204,591 individuals with negative FOBT results. There was a statistically significant relative increase of 11·13% (95% CI 4·51–18·66%) in outpatient visits for the selected diagnoses among individuals with positive FOBT results compared with those with negative results in the post-screening period. This increase was driven primarily by specialist outpatient visits in gastrointestinal care (86·42% [95% CI 61·36–120·69%]), and gastroscopy procedures (36·68% [95% CI 13·39–71·99%]). Notably, the analysis did not show changes in primary care utilisation for diagnoses of gastrointestinal diseases or psychiatric health disorders, nor for dispensed prescription drugs.Interpretation: The findings of this study suggest that the consequences of organised CRC screening extend beyond the intended screening pathway. False positive CRC screening results appear to generate downstream healthcare utilisation concentrated within specialist gastrointestinal care, without evidence of spillover to primary care or increased prescription drug use.

Background: Primary care clinicians often work under time pressure and diagnostic uncertainty while managing undifferentiated and evolving symptoms. Patient safety incidents in this setting may arise through interacting clinical, organisational, and communication-related factors. However, little is known about how reported incidents are interpreted locally and translated into organisational improvement. Aim: To examine reported patient safety incidents in Swedish primary care, identify contributory factors, and explore how incident analysis can support organisational learning and patient safety improvement. Design & setting: Mixed-methods study in Region Stockholm, Sweden, using incident reports from primary care centres within a tax-funded universal healthcare system. Method: Descriptive statistics summarised key variables from structured incident review forms. Free-text data on contributory factors and organisational learning were analysed using reflexive thematic analysis following Braun and Clarke’s six-phase approach. Results: A total of 696 incident reports from 34 centres were included. Most incidents were judged avoidable (71%), and 37% involved diagnostic delay. Five groups of contributory factors were identified: staff-related factors, patient-related factors, work environment, administrative processes, and coordination across care boundaries. Reported learning and improvement actions largely mirrored these factors and included continuing education and supervision, clearer follow-up routines, improved continuity and staffing, enhanced safety-netting and patient involvement, and more timely communication between providers. Conclusion: Incident reports highlight recurring vulnerabilities in follow-up, communication, and organisational processes in primary care. Regular local review can support organisational learning and inform practical measures to improve patient safety in primary care.

Objective: Anaemia is common in healthcare and may indicate undiagnosed cancer. Despite this, evidence regarding risk estimates informing clinical decision-making remains limited, particularly regarding haemoglobin dynamics and the role of mean corpuscular volume (MCV). We aimed to quantify the 18-month risks of incident cancer (IC) and all-cause mortality (ACM) following incident anaemia (IA), and to examine how MCV modifies these risks. Methods and analysis: Population-based, age- and sex-matched cohort study. The study used the Stockholm Early Detection of Cancer Study (STEADY-CAN), covering almost all adults residing in Stockholm County, Sweden, during 2011–2021. STEADY-CAN links laboratory tests with national registers, capturing healthcare use, diagnoses, cancer outcomes and prescribed medications. We included 190 057 adults with IA and 190 057 age- and sex-matched non-anaemic controls from the STEADY-CAN cohort. Eligible individuals were ≥18 years old, cancer-free, had ≥2 Hb measurements during 2011–2020 and a concurrent MCV value at the IA date. IA was defined as the first Hb value from 2012 onwards below 130 g/L in men or 120 g/L in women after prior normal values. Sex-stratified adjusted piecewise competing-risks multi-state Cox regression models were used, with separate HRs for IA during 0–3, 3–6, 6–12 and 12–18 months of follow-up. The main outcome measures were IC and ACM during the 18-month follow-up. Results: IC occurred in 6.2% of male and 2.8% of female IA cases, compared with 2.4% and 1.1% of controls. Corresponding ACM rates were 7.4% and 4.0% in IA cases versus 2.5% and 1.7% in controls. IA implied a 9.17-fold higher IC risk and 8.50-fold higher ACM risk among men during 0–3 months of follow-up, with 8.25- and 6.14-fold higher risks among women (all p<0.001). These risks decreased over time but were still significant for both sexes at 6–12 months of follow-up for IC and 12–18 months of follow-up for ACM. Microcytosis was linked to the highest IC risk, particularly for digestive and haematological cancers, whereas macrocytosis was more strongly associated with ACM. Conclusions: IA is a strong marker of both IC and ACM in routine care. Microcytic anaemia should prompt timely gastrointestinal evaluation, while macrocytic anaemia warrants broader assessments for comorbid conditions and systemic disease. Persistently elevated risks underscore the need for structured safety-netting and continued follow-up after IA, even without a cancer identification. Our findings highlight the value of using anaemia patterns and MCV in early risk stratification.

Detection of colorectal cancer (CRC) is mainly achieved by clinical assessment. As new treatments become available for metastatic CRC (MCRC), it is important to accurately identify these patients. To develop a predictive model for identifying MCRC in primary health care patients using diagnostic data analysed with machine learning. A case-control study utilising data on primary health care visits for 146 patients >18 years old diagnosed with MCRC in the Västra Götaland Region, Sweden during 2011, and 577 sex-, age, and primary health care centre-matched controls. Stochastic gradient boosting was used to construct a model for predicting the presence of MCRC based on diagnostic codes from primary health care consultations during the year before index (diagnosis) date and number of consultations. Variable importance was estimated using the normalised relative influence (NRI) score. Risks of having MCRC were calculated using odds ratios of marginal effects (OR_ME). The optimal model included 76 variables with non-zero influence, had an area under the curve of 76.5%, a sensitivity of 77.8%, and a specificity of 69.2%. The 10 most important variables had a combined NRI of 61.0%. Number of consultations during the year before index date had the highest NRI at 19.2%, with an OR_ME of 3.3. A machine learning method based on primary health care consultation frequency and diagnoses may be used to identify important variables for predicting presence of MCRC. Both primary health care consultations and associated diagnostic codes need to be taken into consideration.

Early cancer detection is critical for improving survival rates, yet it requires a careful balance between minimising missed diagnoses and avoiding over- investigation. Primary care plays a central role in this process. This thesis aims to enhance understanding of the complexities involved in early cancer detection by analysing symptoms and signs, with the goal of contributing to the development of risk assessment and prediction tools to identify cancer at an early stage within primary care.This thesis is based on five quantitative studies conducted within the Swedish healthcare system. Study I examined symptoms reported by referred patients via questionnaires at the Department of Pulmonary Medicine at Karolinska University Hospital. It investigated whether machine learning could predict which patients subsequently received a lung cancer diagnosis, stratified by smoking status. Studies II and III focused on comprehensive diagnostic data and coded symptoms from primary care to facilitate early detection of non-metastatic colorectal cancer, with Study II conducted in Region Stockholm and Study III in Region Västra Götaland. Studies IV and V used comprehensive clinical and laboratory data from the entire adult population of Stockholm County. Study IV presents a cohort description, while Study V examines the association and discriminatory capacity of newly developed anaemia as an indicator for cancer.In Study I, the findings demonstrate that predictive models, using machine learning, exhibit good discriminatory ability for patients who either never smoked or were current smokers. In Study II, an existing Swedish risk assessment tool was validated by replicating it in a different region, with consistent results across regions. In Study III, a new predictive model was developed using machine learning to analyse all diagnostic data for identifying non-metastatic colorectal cancer. Study IV introduces the extensive STEADY-CAN (Stockholm early detection of cancer study) cohort, providing opportunities to analyse early cancer detection patterns. Study V investigates the association between newly developed anaemia and cancer risk within the STEADY-CAN cohort, revealing a significant impact on risk assessment.These findings collectively address the value of improved risk stratification in primary care by leveraging existing data to better identify patients at elevated risk of having cancer. The results highlight key themes in early detection- predictive accuracy, risk stratification, clinical utility, and applicability across populations-and identify areas where current evidence has been limited or inconclusive. Future research should prioritise the validation of novel diagnostic approaches and the development of systems that support clinical decision- making, while upholding the principles of accessible, patient-centred care. Keywords: Early cancer detection, Primary care, Machine learning, Colorectal cancer, Lung cancer, Risk assessment tools, Anaemia, STEADY-CAN.List of scientific papersI. Nemlander E, Rosenblad A, Abedi E, Ekman S, Hasselström J, Eriksson LE, Carlsson AC. Lung cancer prediction using machine learning on data from a symptom e-questionnaire for never smokers, formers smokers and current smokers. PLoS One. 2022 Oct 21;17(10):e0276703. PMID: 36269746; PMCID: PMC9586380.Erratum for: PLoS One. 2022 Oct 21;17(10):e0276703. PMID: 38060550; PMCID: PMC10703334. https://doi.org/10.1371/journal.pone.0276703II. Nemlander E, Rosenblad A, Abedi E, Hasselström J, Sjövall A, Carlsson AC, Ewing M. Validation of a diagnostic prediction tool for colorectal cancer: a case-control replication study. Fam Pract. 2023 Jan 5:cmac147. PMID: 36611019. https://doi.org/10.1093/fampra/cmac147III. Nemlander E, Ewing M, Abedi E, Hasselström J, Sjövall A, Carlsson AC, Rosenblad A. A machine learning tool for identifying non- metastatic colorectal cancer in primary care. European Journal of Cancer. 2023 Jan 11. https://doi.org/10.1016/j.ejca.2023.01.011IV. Nemlander E, Abedi E, Ljungman P, Hasselström J, Carlsson AC, Rosenblad A. The Stockholm Early Detection of Cancer Study (STEADY-CAN): rationale, design, data collection, and baseline characteristics for 2.7 million participants. https://doi.org/10.1007/s10654-024-01192-8V. Nemlander E, Rosenblad A, Abedi E, Hasselström J, Ljungman P, Carlsson AC. Newly developed anaemia predicts incident cancer and death within 18 months: Findings from 1.1 million patients in the Stockholm Early Detection of Cancer Study (STEADY-CAN) cohort. [Manuscript]

Early cancer detection is critical for improving survival rates, yet it requires a careful balance between minimising missed diagnoses and avoiding over- investigation. Primary care plays a central role in this process. This thesis aims to enhance understanding of the complexities involved in early cancer detection by analysing symptoms and signs, with the goal of contributing to the development of risk assessment and prediction tools to identify cancer at an early stage within primary care.This thesis is based on five quantitative studies conducted within the Swedish healthcare system. Study I examined symptoms reported by referred patients via questionnaires at the Department of Pulmonary Medicine at Karolinska University Hospital. It investigated whether machine learning could predict which patients subsequently received a lung cancer diagnosis, stratified by smoking status. Studies II and III focused on comprehensive diagnostic data and coded symptoms from primary care to facilitate early detection of non-metastatic colorectal cancer, with Study II conducted in Region Stockholm and Study III in Region Västra Götaland. Studies IV and V used comprehensive clinical and laboratory data from the entire adult population of Stockholm County. Study IV presents a cohort description, while Study V examines the association and discriminatory capacity of newly developed anaemia as an indicator for cancer.In Study I, the findings demonstrate that predictive models, using machine learning, exhibit good discriminatory ability for patients who either never smoked or were current smokers. In Study II, an existing Swedish risk assessment tool was validated by replicating it in a different region, with consistent results across regions. In Study III, a new predictive model was developed using machine learning to analyse all diagnostic data for identifying non-metastatic colorectal cancer. Study IV introduces the extensive STEADY-CAN (Stockholm early detection of cancer study) cohort, providing opportunities to analyse early cancer detection patterns. Study V investigates the association between newly developed anaemia and cancer risk within the STEADY-CAN cohort, revealing a significant impact on risk assessment.These findings collectively address the value of improved risk stratification in primary care by leveraging existing data to better identify patients at elevated risk of having cancer. The results highlight key themes in early detection- predictive accuracy, risk stratification, clinical utility, and applicability across populations-and identify areas where current evidence has been limited or inconclusive. Future research should prioritise the validation of novel diagnostic approaches and the development of systems that support clinical decision- making, while upholding the principles of accessible, patient-centred care. Keywords: Early cancer detection, Primary care, Machine learning, Colorectal cancer, Lung cancer, Risk assessment tools, Anaemia, STEADY-CAN.List of scientific papersI. Nemlander E, Rosenblad A, Abedi E, Ekman S, Hasselström J, Eriksson LE, Carlsson AC. Lung cancer prediction using machine learning on data from a symptom e-questionnaire for never smokers, formers smokers and current smokers. PLoS One. 2022 Oct 21;17(10):e0276703. PMID: 36269746; PMCID: PMC9586380.Erratum for: PLoS One. 2022 Oct 21;17(10):e0276703. PMID: 38060550; PMCID: PMC10703334. https://doi.org/10.1371/journal.pone.0276703II. Nemlander E, Rosenblad A, Abedi E, Hasselström J, Sjövall A, Carlsson AC, Ewing M. Validation of a diagnostic prediction tool for colorectal cancer: a case-control replication study. Fam Pract. 2023 Jan 5:cmac147. PMID: 36611019. https://doi.org/10.1093/fampra/cmac147III. Nemlander E, Ewing M, Abedi E, Hasselström J, Sjövall A, Carlsson AC, Rosenblad A. A machine learning tool for identifying non- metastatic colorectal cancer in primary care. European Journal of Cancer. 2023 Jan 11. https://doi.org/10.1016/j.ejca.2023.01.011IV. Nemlander E, Abedi E, Ljungman P, Hasselström J, Carlsson AC, Rosenblad A. The Stockholm Early Detection of Cancer Study (STEADY-CAN): rationale, design, data collection, and baseline characteristics for 2.7 million participants. https://doi.org/10.1007/s10654-024-01192-8V. Nemlander E, Rosenblad A, Abedi E, Hasselström J, Ljungman P, Carlsson AC. Newly developed anaemia predicts incident cancer and death within 18 months: Findings from 1.1 million patients in the Stockholm Early Detection of Cancer Study (STEADY-CAN) cohort. [Manuscript]

The Stockholm Early Detection of Cancer Study (STEADY-CAN) cohort was established to investigate strategies for early cancer detection in a population-based context within Stockholm County, the capital region of Sweden. Utilising real-world data to explore cancer-related healthcare patterns and outcomes, the cohort links extensive clinical and laboratory data from both inpatient and outpatient care in the region. The dataset includes demographic information, detailed diagnostic codes, laboratory results, prescribed medications, and healthcare utilisation data. Since its inception, STEADY-CAN has collected longitudinal data on 2,732,005 individuals aged ≥ 18 years old living in or having access to health care in Stockholm County during the years 2011–2021. Focusing on cancer, the cohort includes 140,042 (5.1%) individuals with incident cancer and a control group of 2,591,963 (94.9%) cancer-free individuals. The cohort’s diverse adult population enables robust analyses of early symptom detection, incidental findings, and the impact of comorbidities on cancer diagnoses. Utilizing the wide range of available laboratory data and clinical variables allow for advanced statistical analyses and adjustments for important confounding factors. The cohort’s primary focus is to improve understanding of the early diagnostic phase of cancer, offering a crucial resource for studying cancer detection in clinical practice. Its comprehensive data collection provides unique opportunities for research into comorbidities and cancer outcomes, making the cohort a useful resource for ongoing cancer surveillance and public health strategies. The present study gives a detailed description of the rationale for creating the STEADY-CAN cohort, its design, the data collection procedure, and baseline characteristics of collected data.

Background: Suspected malignant skin lesions are common and challenging in terms of diagnostic accuracy and lead time. Efficient use of teledermatoscopy has the potential to improve these factors.Objectives: This report aims to explore the implementation of teledermatoscopy for quality improvement in a comprehensive healthcare system.Methods: A stepwise implementation started in 2012 in the Stockholm Region, Sweden, serving 2.4 million inhabitants across approximately 240 primary healthcare centres. The full‐scale implementation occurred from May 2021 to February 2023, utilising a double‐reading teledermatoscopy system. The RE‐AIM framework (reach, effectiveness, adoption, implementation, maintenance) guided the evaluation, including measures such as participant engagement, referrals, diagnoses, staff participation, and centres going on post‐implementation. Descriptive statistics were used.Results: In the full‐scale implementation period, 10,862 referrals were transmitted, with 17% recommended for biopsy or excision. Among these, 282 (15%) were confirmed melanomas and 516 (28%) non‐melanoma skin cancers. Challenges were observed during the stepwise phase, particularly in engaging primary healthcare and establishing dermatologic consultation capacity. However, all centres and 1263 primary care physicians participated in the fully scaled implementation. Challenges included contractual compliance, adopting a collective system perspective, effective management transition of referrals, cost allocation, and communication improvements. Six months post‐study, most centres used the system, with around 1500 monthly referrals.Conclusions: The evaluation revealed the overall satisfactory performance of the teledermatoscopy system in terms of reach, effectiveness, adoption, implementation, and maintenance during the fully scaled phase. Results underscore the effectiveness of a stepwise implementation to successfully address contextual challenges. The findings emphasise the system's potential for a broader patient outreach. In conclusion, implementation of teledermatoscopy for quality improvement in a comprehensive healthcare system is an organisational challenge but has the potential to improve healthcare quality for suspected malignant skin lesions.