This systematic review of evidence was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.31 The review has been registered on the International Prospective Register of Systematic Reviews (PROSPERO).32 The protocol, which was designed in accordance with the PRISMA Protocols guidelines,33 has been published elsewhere.34

The following databases were searched: MEDLINE (via PubMed), Cochrane Library (including the Cochrane Central Register of Controlled Trials (CENTRAL) and the Database of Abstracts of Reviews of Effects (DARE)), Scopus, CINAHL, Web of Science and the International Clinical Trials Registry Platform (ICTRP). Relevant grey literature sources, including the publication database of the WHO’s International Agency for Research on Cancer, the Cancer Atlas of the Union for International Cancer Control and the Global Cancer Project Map, were also searched for potentially eligible publications. The search strategy (see online supplemental appendix 1) was developed with guidance from a health science subject librarian. More details of the search strategies used are described in the published review protocol.34 Only articles published over the last 10 years (from 1 January 2010 to the date of search) were considered eligible, with no language restriction. In addition, the bibliographical references of included studies were also searched. Search was conducted from 22 June 2020 through 30 November 2020.

The study inclusion criteria were guided by the research question: ‘What interventions targeting the key steps outlined by the WHO operational framework for early diagnosis have been used for improving timely diagnosis of symptomatic breast and cervical cancers in LMICs, and how effective are they?’. Eligibility was determined the PICO (participants; interventions; comparator; outcomes) criteria in line with the research question. Participants included women, general public or healthcare workers (HCW) in LMICs. The definition of LMICs was based on the World Bank’s current classification using per capita gross national income.35 Interventions included strategies aimed at influencing the timeliness of symptomatic breast or cervical cancer diagnosis along any or a combination of the three steps of the WHO framework for early diagnosis, whether as a single focus intervention or as a multifocus intervention targeting more than one cancer type. Studies with interventions focused primarily on screening of asymptomatic individuals were excluded. Comparator was the standard of care. Outcomes were those related to any or a combination of the three steps of the WHO framework for early diagnosis, such as improved access to early diagnostic services, reduced diagnostic time and tumour downstaging. More detailed information about the eligibility criteria can be found in the published review protocol.34

The review consisted of two levels of screening: a title and abstract screening to identify potentially eligible publications and review of full texts to select those to be included in the review based on predefined inclusion/exclusion criteria. For the first level of screening, two reviewers (CAN and PK) independently screened the titles and abstracts of all retrieved records from the search output. Articles that were considered relevant by either or both of the reviewers were included in the full-text review. In the second step, the two reviewers (CAN and PK) independently assess the full texts to determine whether they met the inclusion/exclusion criteria. Any discordance in eligibility assessment was resolved through consensus between the two researchers, with recourse to a third arbitrator (JM) as necessary.

Two reviewers (CAN and PK) independently extracted all relevant data from the included articles using a standardised data extraction tool, adapted from the framework proposed by Carlos and colleagues.36 Extracted data fell under four domains: (1) study identification details (article title, journal title, authors, country of the study, language, publication year, host institution of the study), (2) methodological characteristics (study design, study objective or research question or hypothesis, sample characteristics (eg, sample size; sex; age, ethnicity; groups and controls; follow-up duration; validation of measures; statistical analyses), (3) main findings and (4) conclusions. Study eligibility was reverified at the start of data extraction. Discrepancies in extraction were resolved through consensus. CAN combined the two spreadsheets of extracted data for analysis, and data were double checked by PK for completeness and accuracy.

Two reviewers (CAN and PK) independently assessed each included study for risk of bias, using the Cochrane risk of bias tool37 for randomised clinical trials (RCT), and the Risk of Bias In Non-randomised Studies of Interventions tool38 for non-randomised and observational studies. Discrepancies in study quality assessments were resolved by discussion and consensus. Further details of the quality appraisal process are discussed in the published protocol of this review.34

A PRISMA flow diagram was used to illustrate the literature search results and study selection process. Outcomes were categorised according to the essential steps of early cancer diagnosis as specified in the WHO Guide to Cancer Early Diagnosis.24 Plans to conduct a meta-analysis were not feasible as studies were not quantitatively comparable in terms of interventions and reported outcomes. Instead, a thematic narrative synthesis approach was used to aggregate and analyse relevant findings. Plans to assess the certainty of reviewed evidence using the Grades of Recommendation Assessment, Development and Evaluation approach39 were not feasible for similar reasons. A table summarising findings from each included study was presented. JM and FMW reviewed the analysed data for accuracy and consistency with protocol.

This is a review of publicly available literature. Patients and the public were not directly involved in the design, conduct, reporting or dissemination this study.

A total of 10 593 records were identified from literature database searches, from which 1244 duplicates were discarded. The remainder, 9349 unique records, had their titles and abstracts screened, from which 9264 clearly ineligible publications were excluded. The full texts of the remaining 85 potentially eligible studies were reviewed against predefined inclusion and exclusion criteria; 21 were included in the review and 64 were excluded for various reasons. Figure 1 presents the PRISMA flow chart of the study selection process and reasons for exclusion.

Tables 1 and 2 summarise the characteristics of included studies, published between 2015 and 2020. Studies were conducted in 20 LMICs across regions of Africa (8), Europe (4), Latin America (2), Middle-East Asia (1) and South-East Asia (5). Twenty were single-country studies, in which 13 countries were represented: Bangladesh (2),8 40 Botswana (2),41 42 Colombia (1),43 India (1),44 Indonesia (1),45 Iran (1),46 Malaysia (2),47 48 Nigeria (1),49 Pakistan (1),50 Rwanda (1),51 South Africa (3),52–54 Tanzania (1)55 and Zambia (3).56–58 One study had a multicountry focus, with participants recruited from nine countries: Bosnia-Herzegovina, Costa Rica, Egypt, India, North Macedonia, Pakistan, Slovenia, Turkey and Uganda.59 Figure 2 illustrates the geographical distribution of the studies.

In terms of study design, six studies were RCT,8 40 43 45 46 51 six were before-and-after studies,42 44 48 49 52 54 one was a longitudinal (cohort-type) observational study,55 four were retrospective cohort-type studies,41 47 50 53 while the remaining four were cross-sectional studies.56–59

The majority (16/21) of included studies primarily focused on interventions addressing breast cancers; two focused on cervical cancer, whereas the rest examined multiple cancer types (including breast and/or cervical cancers). Of the 21 studies, 12 targeted HCWs (including physicians, nurses, pathologists, non-physician health providers and community health workers (CHW)); 7 targeted women and adolescent girls and 3 targeted both women and HCWs. There was an even representation across urban and rural implementation contexts, with reported interventions implemented in 11 urban, 9 rural and 1 mixed (urban and rural) communities.

When categorised by the three steps of the WHO Guide to Cancer Early Diagnosis,24 eight studies reported on interventions aiming to address barriers in step 1 (promoting awareness and addressing delays in access to care)8 40 44–47 49 54; another eight focused on interventions addressing step 2 (addressing diagnostic delay by optimising clinical evaluation, diagnosis and staging)42 43 50–52 55 59 60; one study reported on an intervention targeted at both steps 1 and 253; while five studies assessed interventions targeted at all three steps (addressing access, diagnostic and treatment delays).48 56–58 Tables 1 and 2 describe the categorisation of the interventions by the three steps of the WHO Guide to Cancer Early Diagnosis.24

Only two of the included studies were determined to be of high quality (low risk of bias),43 49 seven were adjudged to be of moderate quality (moderate risk of bias)8 40 45 46 51 52 57 while the majority was found to be of low quality (high risk of bias).41 42 44 47 48 50 53–56 58 59 Figure 3A,B illustrate the risk of bias assessment outcomes of the included studies.

Interventions seeking to reduce access delays by promoting awareness, help-seeking and linkage to care were predominantly targeted at women (5),44–47 54 with a few targeted at adolescent girls (1)49 and community/lay health workers (2).8 40

A cluster-randomised trial assessed a self-help intervention for reducing time to diagnosis in Indonesian women with breast cancer symptoms. The intervention consisted of health education and psychoeducation using a narrative strategy, which involved the use of testimonials and storytelling.45 It reduced the time between the first medical consultation and definitive diagnosis among women with breast symptoms by 13.3 days (25.90 in the intervention group vs 39.29 in the control group, p=0.02). Another RCT by Termeh-Zonouzy evaluated the effectiveness of an educational intervention based on fear appeals using the extended parallel process model to improve attitudes, intention and timely breast cancer diagnosis in Iran. It found a significant improvement in attitude and intention (p=0.01 and p=0.001, respectively), but no significant improvement was observed for early breast cancer diagnosis (p=0.78).46

An RCT by Gadgil and colleagues investigated a programme that aimed to increase awareness on breast cancer and access to timely detection by emailing awareness brochures annually to a cohort of women in an urban occupational setting in India.44 They found an increase in the proportion of women with early breast tumours and a concomitant increase in the proportion of women receiving breast conserving surgery increased postintervention when compared with the preintervention period increased from (74% to 81% and 39% to 51%, respectively). However, there were no outcomes directly related to the timeliness of the diagnoses.

A before-and-after study assessed the effectiveness of decentralising colposcopy services from tertiary-level to primary-level care facility in an urban community in South Africa to aid early definitive diagnosis of cervical cancer.54 The number of women who had a colposcopy at the primary healthcare facility rose threefold postdecentralisation (from 114 to 350), with an increased proportion of women who had a colposcopy within 3 months of a Pap smear (from 11.8% to 15.4%) and reduced workload at the tertiary facility. A retrospective study compared routine (opportunistic) screening mammogram with targeted (high risk) screening mammogram and diagnostic mammogram among women in Malaysia.47 It observed cancer detection rates of 0.5 %, 1.25% and 26% by opportunistic screening, targeted screening and diagnostic mammograms, respectively. The proportion of non-invasive cancer (ductal carcinoma in situ) was higher (23.1 %) in the two screening groups compared with only 2.5% in the diagnostic group, while the proportion of invasive cancer was higher in the diagnostic group. None of these outcomes was directly related to the timeliness of diagnosis.

Of the interventions targeting lay/CHW, an RCT by Ginsburg et al assessed the role an mHealth model that used smartphone applications and CHW training as part of a patient navigation programme to address potential barriers to seeking care in a rural community in Bangladesh.8 The intervention was found to have increased clinic attendance for breast symptoms, improved data quality, higher identification of women with abnormal breast examinations and better adherence to diagnostic follow-up. Similarly, another RCT of a CHW-driven population-based case finding programme for breast cancer in a rural setting in Bangladesh found the use of mobile technology as feasible and efficient in active breast cancer case finding in rural settings. Manual data collection and reporting had missing data in 80% of cases, and took an average of 5 min longer to collect data, vs no missing data when mobile phones were used for data collection and reporting.40

A school-based cervical cancer prevention and early diagnosis education programme targeted at rural teenage high school students Nigeria.49 Notably, no significant improvements in knowledge parameters (Pap smears, risk factors and early symptoms). Nevertheless, participants who engaged more with the intervention showed significant improvements in knowledge across most parameters of knowledge.

An RCT compared diagnostic outcomes between a physician-targeted opportunistic clinic‐based breast cancer screening programme and usual care in an urban setting in Colombia.Physicians in the intervention clinics received a 2-day training course on breast cancer epidemiology, clinical signs and symptoms as well as the principles of mammography and Breast Imaging Reporting and Data System grading. They were instructed to perform clinical breast examination (CBE) on all eligible women, record results and refer women with suspicious findings for further diagnostic procedures. It demonstrated a higher proportion of early breast cancer diagnosed with the intervention; 68.2% of the women with breast cancer in the intervention group had breast conservation surgery, compared with 50.0% in the control group.43

A before-and-after study of a breast ultrasound training programme for non-physician providers with the goal of early diagnosis of breast cancer and downstaging in a rural setting in South Africa demonstrated improved diagnostic competencies (with pretest to post-test averages improvement of 68% in total across four competencies) (foundational knowledge, descriptive categories, benign vs malignant and lesion identification).52 No outcome directly related to diagnostic timeliness was reported, however. Another before-and-after study studied district-level capacity building training programme targeting primary care providers in Botswana, aiming to enhance timely diagnosis of cancers. It found an overall performance increase (measured by the immediate impact of training on the acquisition of knowledge, attitude and skills in eight subdomains: pathophysiology, epidemiology, social context, symptoms, evaluation, treatment, documentation and follow-up) of 16.8% after participation across all subdomains.42

A prospective cohort study by Ngoma et al assessed the role of village navigators for case finding and showed significant downstaging of cancers in a Tanzanian village across 3 successive years (p<0.001).55 An observational clinical study examined the effectiveness of a programme aiming to improve early detection of breast cancer by strengthening both the clinical and quality aspects of the breast imaging practice of oncology centres.59 The interventions included a shift to image-guided core biopsies, implementing multidisciplinary breast meetings, implementing quality assurance procedures, transitioning to digital technologies, facilitating training activities for relevant professionals and improving data collection processes. It was observed that the programme enhanced facility-level breast cancer early diagnostic capacity and increased number of mammography (in 7 out of 10 participating centres) and timely biopsies performed (in 8 out of 10 centres).

A retrospective study assessed the relationship between adequate clinical information (CI) and breast cancer histopathological diagnostic turnaround time (TAT) in Pakistan.50 It found that breast tissue specimens with deficient CI were associated with longer TAT (>80% of the specimens took had TAT of longer than 3 days vs <50% of specimens with sufficient CI). Another retrospective study that assessed a pathology scale-up programme to optimise breast cancer pathological diagnostic TAT in Botswana found that the programme decreased median TAT for biopsy and immunohistochemistry specimens (from 21.5 days to 8 days, before and after the initiation of the programme, respectively). However, there was no significant decline in median TAT for surgical specimens.41

A cluster-randomised RCT by Pace et al investigated the impact of a HCW training programme on diagnosis and staging to facilitate breast cancer early diagnosis in a rural district of Rwanda.51 It consisted of training of CHWs in how to educate community members about breast cancer symptoms recognition and early help seeking; training of nurses in symptom recognition, CBE and appropriate referral; training of hospital-based clinicians in diagnostic breast ultrasound and ultrasound-guided core needle biopsy. The intervention led to an increase in diagnostic health facility visits (an increase of 4.7–7.9 visits/month in intervention facilities compared with control facilities, p<0.05) and more breast biopsies (36.6 vs 8.9/100 000 person-years, p<0.001) and higher incidence of early-stage breast cancer (3.3 per 100 000 vs 0.7 per 100 000 in control areas, p<0.05).

Hadley and colleagues conducted a retrospective (postimplementation) study to determine the impact of an improved and computerised breast cancer patient-tracking system and database, as part of a nurse-driven and ultrasound-based breast cancer early detection programme in a rural community in South Africa.53 Most health workers’ involved agreed that the intervention can improve early breast cancer detection, patient follow-up, clinical decision-making, communication among clinicians. However, a majority (71%) of them reported that the system increased visit time, due to factors such as unreliable internet and inadequate computer workstations.

There were interventions targeting all three steps of the WHO guide. A patient navigation programme to improve timely breast cancer diagnosis, treatment and follow-up for women in Malaysia was found to have increased the proportion of women receiving timely mammography (from 74.4% to 96.4%), biopsy (from 76.1% to 92.5%) and timely communication of diagnostic results (from 58.5% to 80.0%).48 Other interventions included the establishment of a district-level breast care specialty clinic that offers same-day breast self-awareness education, clinical breast examination, breast ultrasound, ultrasound-guided breast biopsy, surgery, referral for treatment and follow-up for women with breast symptoms in Zambia58; an algorithm that compresses the multistep breast cancer symptom awareness, diagnostic and treatment pathway into a single clinic visit56 and an initiative that leverages an existing cervical cancer prevention service platform to build facility-level capacity for breast cancer care by raising awareness among women, creating a resource-appropriate training curricula for mid-level and high-level providers for early detection and treatment.57 Due to the cross-sectional nature of the studies reporting these interventions, they could only demonstrate the feasibility of the interventions within resource-limited contexts, and not their effectiveness.

Our review has some limitations worthy of note. First, most of the included studies assessed intervention outcomes using observational, non-experimental and non-standardised methods, making them prone to recall bias, selection bias, uncontrolled confounding and limited external validity. Second, as with other reviews, our study may be susceptible to publication bias in the reporting of study findings, in which positive findings have a higher likelihood of being reported. This may have been reflected in our finding of very limited evidence of non-effectiveness of interventions. Third, only a few of the studies measured intervention outcomes beyond a duration of 1 year, hence, limiting the determination of the long-term impact of reviewed interventions. Furthermore, although we searched a wide array of literature sources, it is possible that we missed some relevant studies. Finally, our study excluded interventions that focused primarily on screening of asymptomatic individuals for reasons earlier alluded to. Hence, it is possible that some of the screening studies excluded had early diagnosis components.