Purpose
Cancer Insight is an independent, educational explorer of cancer research literature. Its job is to show what research exists while keeping the original evidence visible: what was retrieved, where it came from, why it appears and what the numbers do — and do not — mean.
It does not diagnose, recommend treatments or give medical advice. Counts of papers describe research activity, never how well a treatment works.
Sources
- NCBI PubMed and PubMed Central (U.S. National Library of Medicine), via the E-utilities API: record counts, Best Match and newest-first record lists, and metadata (titles, abstracts, authors, journals, dates, publication types, MeSH headings, DOIs and PMC links).
- Cure Cancer With AI research API: a third-party feed of the newest research records for 16 cancer types (lung cancer, breast cancer, brain tumor, colorectal cancer, prostate cancer, pancreatic cancer, liver cancer, kidney cancer, melanoma, ovarian cancer, cervical cancer, bladder cancer, gastric cancer, esophageal cancer, leukemia, lymphoma), including treatment-category tags. Cancer Insight re-checks every feed record and replaces its metadata with PubMed’s where a PubMed ID exists.
- ClinicalTrials.gov (U.S. National Library of Medicine), via its data API version 2: registered clinical studies for each cancer.
- GLOBOCAN 2024 (Global Cancer Observatory, International Agency for Research on Cancer): worldwide estimates of new cases and deaths, used to compare research with burden.
- Wikimedia Commons: openly licensed scientific images.
Searches are sent from Cancer Insight’s server to these services. Cancer Insight is not affiliated with any of them.
From search to PubMed query
Every cancer type is turned into one PubMed query that combines MeSH headings (which include narrower headings) with title/abstract phrases. The same query is used for every count, list and Gap Atlas measure for that cancer, and it is shown wherever results appear. For example, “pancreatic cancer” becomes:
("Pancreatic Neoplasms"[mh] OR "pancreatic cancer"[tiab] OR "pancreatic cancers"[tiab] OR "pancreatic carcinoma"[tiab] OR "pancreatic adenocarcinoma"[tiab] OR "pancreatic ductal adenocarcinoma"[tiab])Cancer Insight recognises 29 cancer types and common synonyms (for example colon → colorectal, NSCLC → non-small cell lung cancer). Other inputs are searched as title/abstract phrases such as “appendix cancer”. Inputs are limited to letters, numbers, spaces and hyphens so they cannot change the structure of a query.
Recent research
- For cancers the feed covers, Cancer Insight requests the 100 newest feed records (cached for up to an hour).
- Each record is enriched with PubMed metadata and checked for relevance. It is kept when the cancer appears in the title, MeSH headings or keywords, or in the abstract of a record the feed tags with that cancer. Records whose title is about a different cancer are dropped. The matching field is shown on each paper as “Why it’s here”.
- Up to 20 records are kept, most relevant first and then newest. If fewer than 8 pass, the newest PubMed records for the query fill the list and are labelled as such. Cancers the feed does not cover use PubMed directly.
- Treatment tags come from the feed. A “stem cell transplant” tag is kept only when the text describes a transplant, so papers about cancer stem cells are not mislabelled.
Because the feed returns the newest records, this list shows what is new. It is not a sample of the whole literature, and statistics about it (for example its publication years) describe these records only. The Papers view can also show PubMed Best Match: PubMed’s own relevance ranking across all years.
Evidence types
Evidence types come from PubMed publication types, using one classifier everywhere on the site. When a record has several types, the first matching row below wins.
| Label shown | PubMed publication types | Family |
|---|---|---|
| Meta-analysis | Meta-Analysis | Evidence synthesis |
| Systematic review | Systematic Review | Evidence synthesis |
| Randomized controlled trial | Randomized Controlled Trial (with phase when indexed) | Primary research |
| Clinical trial | Clinical Trial; Phase I–IV; Controlled, Pragmatic, Equivalence or Adaptive Clinical Trial | Primary research |
| Guideline | Guideline; Practice Guideline; Consensus Development Conference | Guidance |
| Review | Review; Scoping Review | Evidence synthesis |
| Observational study | Observational Study | Primary research |
| Case report | Case Reports | Primary research |
| Trial protocol | Clinical Trial Protocol | Other |
| Editorial, comment or letter | Editorial; Comment; Letter; News | Commentary |
| Journal article | Journal Article only | Design not indexed |
- Randomized controlled trials always count as clinical trials.
- “Design not indexed” is common: many original studies, including laboratory and clinical research, carry only the generic “Journal Article” type. It is not a judgement.
- Records indexed as “Retracted Publication” show a retraction warning everywhere they appear, including exports.
- Evidence types describe what kind of record a paper is. They are not quality or certainty-of-evidence ratings.
PubMed counts
The Overview counts PubMed records matching the cancer query in total, in each of the last 11 publication years, and for each study design below. Each count is a separate PubMed search and links to it.
| Measure | Added to the cancer query |
|---|---|
| Clinical trials (all phases) | AND "clinical trial"[pt] |
| Randomized controlled trials | AND "randomized controlled trial"[pt] |
| Observational studies | AND "observational study"[pt] |
| Case reports | AND "case reports"[pt] |
| Systematic reviews | AND "systematic review"[pt] |
| Meta-analyses | AND "meta-analysis"[pt] |
| Reviews (all types) | AND review[pt] |
| Guidelines | AND (guideline[pt] OR "practice guideline"[pt]) |
| Free full text available | AND "free full text"[sb] |
| A publication year, e.g. 2024 | AND ("2024"[dp] : "2024"[dp]) |
Read these counts with three cautions: PubMed indexes more journals than it used to, so growth over time partly reflects coverage; the current year is incomplete; and publication types are added during indexing, so the newest records are under-counted by design.
Treatment literature
Treatment categories are matched with the PubMed terms below, combined with the cancer query. For a category, Cancer Insight reports the number of records, their share of the cancer’s literature, the study designs among them, how many were published in the last five years and how many have a free full text. Example records are PubMed Best Match.
| Category | PubMed terms |
|---|---|
| Surgery | (surgery[tiab] OR surgical[tiab] OR resection[tiab] OR "surgical procedures, operative"[mh]) |
| Chemotherapy | (chemotherap*[tiab] OR "antineoplastic combined chemotherapy protocols"[mh]) |
| Radiation therapy | (radiotherap*[tiab] OR "radiation therapy"[tiab] OR radiotherapy[mh]) |
| Immunotherapy | (immunotherap*[tiab] OR "immune checkpoint inhibitor*"[tiab] OR immunotherapy[mh]) |
| Targeted therapy | ("targeted therap*"[tiab] OR "molecular targeted therapy"[mh] OR "kinase inhibitor*"[tiab]) |
| Hormone therapy | ("hormone therap*"[tiab] OR "endocrine therap*"[tiab] OR "hormonal therap*"[tiab] OR "androgen deprivation"[tiab] OR "antineoplastic agents, hormonal"[mh]) |
| Stem cell transplant | ("hematopoietic stem cell transplantation"[mh] OR "stem cell transplant*"[tiab] OR "bone marrow transplant*"[tiab]) |
A comparison of two categories compares literature: how much has been published and in which designs. It does not compare effectiveness or safety, and the two lists of example records are not paired. More records, more trials or newer records never mean a better treatment.
Research Gap Atlas — Method 3.0
The atlas asks whether a study design, research theme, access filter or publication period makes up a smaller share of a cancer’s PubMed literature than of oncology as a whole.
share(cancer) = records(cancer AND measure) ÷ records(cancer)
share(oncology) = records(oncology AND measure) ÷ records(oncology)
ratio = share(cancer) ÷ share(oncology)| Reading | Rule |
|---|---|
| Much lower representation | ratio below 0.33 |
| Lower representation (a signal) | ratio below 0.5 |
| Higher representation (shown, never a signal) | ratio above 1.5 |
| Too few records | fewer than 300 records for the cancer, or fewer than 10 matching records expected at the oncology-wide share |
| Expected (population) | a low share of pediatric studies in cancers that occur mainly in adults, or of older-adult studies in cancers that occur mainly in children |
These thresholds are exploratory design choices, not accepted standards. Because the atlas compares shares across the whole index rather than a sample, sampling error does not apply; the uncertainty lies in how well each query captures its concept.
Measures
| Measure | PubMed terms |
|---|---|
| Clinical trials (all phases) | "clinical trial"[pt] |
| Randomized controlled trials | "randomized controlled trial"[pt] |
| Systematic reviews | "systematic review"[pt] |
| Meta-analyses | "meta-analysis"[pt] |
| Quality of life & patient-reported outcomes | ("quality of life"[mh] OR "quality of life"[tiab] OR "patient reported outcome*"[tiab] OR "patient-reported outcome*"[tiab] OR hrqol[tiab]) |
| Toxicity, adverse events & tolerability | ("drug-related side effects and adverse reactions"[mh] OR toxicit*[tiab] OR "adverse event*"[tiab] OR "adverse effect*"[tiab] OR "side effect*"[tiab] OR tolerability[tiab]) |
| Recurrence & relapse | ("neoplasm recurrence, local"[mh] OR recurrence[tiab] OR relapse*[tiab]) |
| Treatment resistance | ("drug resistance, neoplasm"[mh] OR chemoresistan*[tiab] OR radioresistan*[tiab] OR "drug resistance"[tiab] OR "treatment resistance"[tiab] OR "therapy resistance"[tiab] OR refractory[tiab]) |
| Biomarkers | ("biomarkers, tumor"[mh] OR biomarker*[tiab]) |
| Children & adolescents (title/abstract only) | (pediatric*[tiab] OR paediatric*[tiab] OR child*[tiab] OR adolescen*[tiab]) |
| Older adults (title/abstract only) | (elderly[tiab] OR "older adult*"[tiab] OR "older patient*"[tiab] OR geriatric*[tiab] OR "older people"[tiab]) |
| Survivorship & late effects | ("cancer survivors"[mh] OR survivorship[tiab] OR "cancer survivor*"[tiab] OR "late effect*"[tiab]) |
| Access, equity & disparities | ("healthcare disparities"[mh] OR "health status disparities"[mh] OR disparit*[tiab] OR inequit*[tiab] OR "access to care"[tiab] OR underserved[tiab] OR socioeconomic[tiab]) |
| Free full text available | "free full text"[sb] |
| Published in the last three years | ("YYYY−2"[dp] : "YYYY"[dp]) |
Population themes use title/abstract wording only, so they count studies that focus on a population rather than every study that enrolled it. The oncology baseline is:
(neoplasms[mh] OR cancer*[tiab] OR carcinoma*[tiab] OR tumor*[tiab] OR tumour*[tiab] OR neoplasm*[tiab] OR leukemia*[tiab] OR leukaemia*[tiab] OR lymphoma*[tiab] OR melanoma*[tiab] OR sarcoma*[tiab] OR glioma*[tiab] OR myeloma*[tiab])If a live baseline count cannot be retrieved, the atlas uses a dated reference snapshot for that measure and says so.
What the atlas deliberately does not do
- It never flags treatment categories. It reports how many records mention each one, but relevance depends on the biology of each cancer, so a low count is not a gap.
- It never claims that a real scientific or clinical research gap exists.
- It never ranks cancers, treatments or topics by importance.
Evidence behind the atlas
To make the queries inspectable, the atlas also retrieves the PubMed Best Match records inside four publication periods (the last 3 years, the 5 before, the 10 before those and everything earlier), checks them for cancer relevance and lists them with their period, evidence type and metadata completeness. Equal numbers are requested per period, so this sample is spread across time by design and does not feed any signal.
Changes from version 2.0
Version 2.0 computed coverage inside a sample of 28–100 records. Three problems led to Method 3.0:
- Treatment-focused records, deliberately added for the most common treatments, were counted in general study-design and topic denominators, and they made the treatment distribution signal circular.
- The “recent literature” measure was produced by the sampling design: one of the four sampling periods was the recent period.
- Fixed thresholds on small samples could flag large literatures as lacking trials: on 9 October 2026 the published atlas (method 1.0) reported “0 clinical trials” for pancreatic cancer, which has more than 4,000 clinical-trial records in PubMed.
All-cancer map
The map applies the atlas to every cancer in the catalogue (29 cancers) and every atlas measure at once. Each cell is the same ratio as in the atlas — the cancer’s share of records matching the measure divided by the oncology-wide share — and links to the PubMed search behind it. Counts are cached for a day, so the map fills quickly after its first build.
- Colours use a diverging scale centred on a ratio of 1: brown for lower representation, blue for higher, and a pale neutral between 0.8 and 1.25. Every cell also shows its value, so colour is never the only cue. The map is built as a table that screen readers can navigate, and it can be downloaded as CSV.
- Only ratios below 0.5 are signals, exactly as in the atlas. The extra colour steps are there to make the picture readable, not to create new categories.
- The map compares each cancer with oncology overall. Comparing cancers with each other should be done with care: their queries differ in breadth, and a broad query (such as head and neck cancer) includes several diseases.
Clinical trials
Trial figures come from ClinicalTrials.gov (data API version 2). For each cancer one condition search is sent — ClinicalTrials.gov adds its own synonyms — every matching study is downloaded with a small set of fields (identifier, status, phase, study type, whether results are posted, start and completion dates, lead sponsor type and site countries), and every figure is counted from that single list. The list can be downloaded as CSV, one row per study, with the categories used on the page, so any number can be checked. Up to 40,000 studies are counted per cancer; if a search returns more, the page says so.
The cancer name is sent as the condition, except for the cancers below, where the registry’s usual term is used:
| Cancer | Condition searched on ClinicalTrials.gov |
|---|---|
| Non-small cell lung cancer | Non-small Cell Lung Cancer |
| Brain tumor | Brain Tumor |
| Hepatocellular carcinoma | Hepatocellular Carcinoma |
| Skin cancer | Skin Cancer |
| Multiple myeloma | Multiple Myeloma |
The study-type split counts every registered study. Statuses, phases, sponsors, countries and start years count interventional trials only. Statuses are grouped as follows:
| Measure | Definition |
|---|---|
| Open or about to open | Recruiting; Not yet recruiting; Enrolling by invitation |
| Running, not recruiting | Active, not recruiting |
| Completed | Completed |
| Stopped early or withdrawn | Terminated; Withdrawn; Suspended |
| Status not verified recently | Unknown status |
Completed without results posted
This measure counts interventional trials whose status is “Completed”, whose primary completion date (or completion date, if the primary date is missing) is at least 24 months ago, and which have no results section on ClinicalTrials.gov. The share is out of all completed interventional trials past the same date.
- It is not proof that results were never shared. Many trials report in journals or elsewhere instead, and many studies — for example phase 1 trials, or trials outside U.S. law — are not required to post results on ClinicalTrials.gov.
- The 24-month allowance is deliberately longer than the 12 months U.S. law gives most trials it covers, so recently finished trials are not counted.
- Statuses are reported by sponsors and are not always kept up to date; “status not verified recently” is ClinicalTrials.gov’s own label for records it considers out of date.
- Countries are those listed as trial sites. A trial with sites in several countries counts once in each, so country counts add up to more than the number of trials.
Research compared with burden
This comparison sets each cancer’s share of recent research beside its share of deaths (or new cases) worldwide.
research share = records(cancer, last 5 complete years) ÷ records(oncology, same years)
burden share = deaths(cancer) ÷ deaths(all cancers) (or new cases)
ratio = research share ÷ burden share- Burden: GLOBOCAN 2024 estimates — Cancer Today (2024 estimates, version 1.0), World, both sexes, all ages — transcribed from the world fact sheet and checked twice against it. Shares are of all cancers, including non-melanoma skin cancer, as GLOBOCAN reports them.
- Research: each cancer’s PubMed query (the same one used everywhere on the site) limited to the last five complete publication years, divided by the oncology baseline for the same years.
- A ratio above 1 means the cancer’s share of recent research is larger than its share of deaths; below 1, smaller. Research shares overlap — one paper can be about several cancers — so they do not add up to 100%.
- Burden is one reason among many to study a cancer. Survival, curability, how many people live with the disease and scientific opportunity matter too. The ratio describes where attention goes; it never says how much research a cancer should get.
Catalogue cancers are matched to GLOBOCAN sites as follows:
| Cancer | GLOBOCAN 2024 site(s) |
|---|---|
| Lung cancer | Lung |
| Breast cancer | Breast |
| Colorectal cancer | Colorectum |
| Prostate cancer | Prostate |
| Gastric cancer | Stomach |
| Thyroid cancer | Thyroid |
| Liver cancer | Liver |
| Bladder cancer | Bladder |
| Cervical cancer | Cervix uteri |
| Lymphoma | Non-Hodgkin lymphoma + Hodgkin lymphoma |
| Pancreatic cancer | Pancreas |
| Leukemia | Leukaemia |
| Esophageal cancer | Oesophagus |
| Kidney cancer | Kidney |
| Endometrial cancer | Corpus uteri |
| Melanoma | Melanoma |
| Ovarian cancer | Ovary |
| Brain tumor | Brain, central nervous system |
| Multiple myeloma | Multiple myeloma |
| Testicular cancer | Testis |
| Mesothelioma | Mesothelioma |
| Head and neck cancer | Lip, oral cavity + Larynx + Nasopharynx + Oropharynx + Hypopharynx + Salivary glands |
These catalogue entries are not compared:
| Cancer | Why it is not shown |
|---|---|
| Non-small cell lung cancer | a subtype of a cancer listed above; GLOBOCAN does not report it separately |
| Hepatocellular carcinoma | a subtype of a cancer listed above; GLOBOCAN does not report it separately |
| Renal cell carcinoma | a subtype of a cancer listed above; GLOBOCAN does not report it separately |
| Glioblastoma | a subtype of a cancer listed above; GLOBOCAN does not report it separately |
| Skin cancer | GLOBOCAN reports melanoma and non-melanoma skin cancer under different definitions |
| Sarcoma | not reported as a separate site in GLOBOCAN |
| Neuroblastoma | not reported as a separate site in GLOBOCAN |
Citation: Ervik M, Laversanne M, Lam F, Colombet M, Ferlay J, Filho AM, Bray F (2026). Global Cancer Observatory: Cancer Today (2024 estimates, version 1.0). Lyon, France: International Agency for Research on Cancer. Available from: https://gco.iarc.who.int/today
Retractions
Retracted records are those PubMed indexes with the publication type “Retracted Publication”:
(cancer query) AND "retracted publication"[pt]Cancer profiles report how many of a cancer’s records are retracted, the rate per 10,000 records, and the most recently published retracted records. Retractions reach PubMed after the journal publishes a notice, so the count lags; and a retraction can be for honest error as well as misconduct. Papers lists and exports carry a retraction warning wherever such a record appears.
Cancer profiles
Each catalogue cancer has a profile page that gathers, on one page, the figures described above: its PubMed literature, its row of the all-cancer map, its clinical trials, its retractions and, where GLOBOCAN reports it, its burden. Every figure is computed by the same code as the full tool it comes from and links to it.
Images
Images come from Wikimedia Commons searches for the cancer combined with histopathology, micrograph, CT and MRI. Documents, scans of books, posters and awareness campaigns are excluded; each image needs the cancer in its title, or in its description together with a specific imaging term. Creator, licence and a licence link are shown. Images are educational illustrations, not diagnostic references.
Exports and saved papers
- CSV files start with provenance rows (date, cancer, query, source, filters) and neutralise cells that spreadsheets could run as formulas.
- RIS files import into Zotero, EndNote and Mendeley; retractions are noted.
- PDF reports state when and how records were retrieved, the query, the evidence-type mix and links to every source.
- Gap Atlas JSON contains every count, query, link, threshold and sampled record.
- Saved papers are stored only in your browser, with enough metadata to show them across searches.
Open data
The figures behind every analysis can be fetched as JSON (and some as CSV) from stable addresses listed on the open data page. Each response includes the searches that produced it. The data keep the terms of their original sources; burden figures in particular are IARC’s and should be cited to IARC.
English and Arabic
The whole site is available in Arabic at /ar, laid out right to left. Searches in Arabic are matched to the same catalogue, so the same PubMed and ClinicalTrials.gov queries — which are always in English — produce the same numbers in both languages. Paper titles and abstracts are shown in their original language, and exported files stay in English so they work in reference managers. Numbers use Western digits in both languages.
Reproducibility
PubMed counts and ClinicalTrials.gov lists are cached for up to 24 hours and record lists for up to 12 hours, so a number can lag its source by a day. Opening a count’s link runs the same query on PubMed; small differences are normal as records are added and re-indexed. Requests follow NCBI’s usage policy (identified tool, contact address, rate limits and retries).
Limitations
- PubMed does not index every journal, conference abstract, trial registry or preprint.
- MeSH headings and publication types lag behind the newest records.
- Title/abstract matching catches incidental mentions and misses unusual wording.
- The research feed’s coverage and tagging rules are controlled by its provider.
- Deduplication relies on PubMed IDs, DOIs and titles and can miss duplicates.
- ClinicalTrials.gov is one registry among several; trials registered only elsewhere (for example in the EU or WHO registries) are not counted.
- GLOBOCAN figures are modelled estimates; for countries with limited cancer registration they carry wide uncertainty.
If something looks wrong, tell us — include the cancer, the page and, if possible, the PubMed ID.
Version history
- 11 October 2026 — clinical trials from ClinicalTrials.gov; research compared with GLOBOCAN 2024 burden; all-cancer map; cancer profiles; retraction counts; open data; Arabic edition. The head and neck cancer query was narrowed to mouth, otorhinolaryngologic and salivary gland headings, so its counts are lower than before.
- 9 October 2026 — Research Gap Atlas Method 3.0; shared query catalogue; PubMed counts with linked queries; evidence-type classifier; retraction warnings.
- Method 2.0 (not released) — year-stratified sample and provenance export, superseded by 3.0 before release.
- Method 1.0 — first Research Gap Atlas, based on the records retrieved for a search.