Bacterial Infections and Cancer Remissions

Epistemic status: pretty confident but only making weak claims

In previous posts I’ve talked about how progress on cancer treatment has slowed and how recent targeted chemotherapy drugs are mostly ineffective, and floated some heuristics as to how we might approach cancer research better.  In this and subsequent posts I’ll look at some examples of therapies and research directions that I think are promising. Mostly these aren’t of the form “this is a cure for cancer, go out and take it right now” so much as they are “there might be a cure for cancer around this vicinity, it needs further research, and it seems underappreciated.”  I don’t claim to have an exhaustive list, but I do think I have enough examples to convince a reader that there’s a lot more interesting stuff out there than the conventional “cancer is hard because the low-hanging fruit is gone” story would imply.

In this post, I want to talk about the evidence that bacterial infections sometimes cause complete remissions in cancers.

Antitumor Immunity

Since the 1980s, the scientific consensus has converged on the view that the immune system destroys cancerous cells as they arise. A clinical case of cancer is simply one in which the body fails to fight off the tumor before it becomes large enough to be measurable or cause symptoms.

The current model of antitumor immunity works as follows:

Here’s how antitumor immunity is currently believed[1] to work:

  • dendritic cells sample antigens from the tumor
  • dendritic cells receive a signal to mature and differentiate
  • these activated dendritic cells generate T-cell responses in the lymphatic tissues
  • cytotoxic T cells enter the tumor to perform their function

So activating either dendritic cells or T-cells could result in a stronger immune response to cancer. This is the theory behind many experimental immunotherapies, including cancer vaccines and CAR T-cell immunotherapy.[2]

Spontaneous Remission and Regression

Sometimes cancer regresses or disappears by itself, without treatment; very frequently these cases follow a feverish infection.  There have been many anecdotal and historical accounts of such “miraculous” recoveries, including the story of St. Peregrine.[3]

A review [4] of case studies found 237 cases of spontaneous regression in 1900-1965, and 504 cases from 1966-1987.  These occurred in many kinds of solid and hematological cancers. The most common kinds were kidney, neuroblastoma, malignant melanoma, choriocarcinoma, and bladder.  These are not the commonest cancers; lung, colon, and breast cancer were underrepresented among spontaneous regressions.  Another review [5] observed that “the prevalent view regarding the mechanism for spontaneous regression is the involvement of immunological factors in the host,” noting that regressing tumors have been observed to have elevated cytokine counts or elevated levels of cytotoxic lymphocytes.  A 1976 survey [8] concluded, “Spontaneous remission of acute leukemia is associated with bacterial infection and is of short duration, weeks to months. Spontaneous regression of lymphoma or plasma cell dyscrasia is often of substantial duration, months or years, and frequently is associated with viral infections.” Another review article [10] confirms the relationship between bacterial infection and spontaneous remission in leukemia: “In 1950, Shear reported that brief remissions in children with untreated leukemia were observed in about 10% of the patients. Three quarters of these remissions were preceded by an episode of acute infection.”

A case study [6] from 1976 of a patient with malignant melanoma who refused treatment observed a spontaneous remission; there was an inflammatory reaction with lymphocytic infiltrate and increased lymphocyte cytotoxicity over the time course of regression, consistent with an immunological cause.

A case study [7] of acute myeloblastic leukemia found a spontaneous remission; the remission occurred after a severe febrile pneumonia, which was treated with leukocyte transfusions.

A case study [9] of two patients with acute myeloblastic leukemia observed that both had spontaneous remissions after infection and blood transfusions.

A case study [11] of a patient with acute myeloid leukemia observed a spontaneous remission after a severe streptococcal infection.

A case study [12] of a patient with hepatocellular carcinoma observed a spontaneous remission following massive gastrointestinal hemorrhage, shock, and blood transfusion.

A case series [13] of 52 patients observed that postoperative empyema (collection of pus in the body) after pneumoectomy increases five-year survival in lung cancer from 18% to 50%.

Finally, it is a traditional observation that a history of infectious disease anticorrelates with cancer risk.  A literature review [14] found that cancer patients are less likely than healthy patients to have a history of febrile infections.  “Individuals who had never experienced a febrile infectious disease were 2.5–46.2 times more likely to have developed cancer than those who had had febrile infections.”

This body of evidence suggests that the immune response to infectious disease has an antitumor effect.

Coley Toxins

In 1891, surgeon William Coley treated a patient with a sarcoma by injecting Streptococcus pyogenes culture into the tumor, and caused a complete regression and 8-year remission, after a severe erysipelas infection from which the patient almost died. Heat-killed Streptococci were safer, but didn’t have a tumor-shrinking effect.  Adding heat-killed Serratia marrescens to the mixture made it effective again, and caused 60 out of 210 (29%) terminally ill sarcoma patients to have relapse-free survival of more than 10 years.[10]   By comparison, modern 5-year survival rates [23] for soft-tissue sarcoma are 83% for localized sarcoma, 54% for regional sarcoma, and 16% for sarcoma with distant spread. So Coley’s patients may have had better or comparable results to modern patients.

Coley tried at least 13 different vaccine formulations over time, making it difficult to compare results.  “Although Coley, in his numerous publications, seldom gave full details of site, dosage, frequency, or duration of vaccine application, the optimal therapy regimen, with hindsight, seemed to be intratumoral, intramuscular, or intraperitoneal injections over long periods of time.”[10]

Coley selected his patients deliberately to maximize the chance of recovery; he favored sarcoma patients because the bacterial vaccine seemed to work best on them. Cases of spontaneous remission following erysipelas infections were also predominantly soft-tissue sarcomas.[16]

Coley’s preparations, known as “Coley Toxins”, were in use until 1963, when the Estes-Kefauver Act ruled that drugs had to prove safety and efficacy to be FDA-approved.[15]

A retrospective review [17] published in in Alternative Therapies in Health and Medicine compared 128 Coley cases from 1890 to 1960 with 1675 controls from SEER who received a cancer diagnosis in 1983. Patients were matched by age, site, stage, and radiation treatment status.  10-year survival rate was not significantly different for Coley’s patients vs. modern controls, which suggests that Coley toxins were of comparable efficacy to chemotherapy.

A 1971 retrospective study[18] of 47 patients with reticulum sarcoma of the bone, treated with Coley toxins before 1956, found a five-year survival rate of 64%.  22 cases (48%) achieved permanent results.  With surgery or radiation therapy alone, five-year survival rates were 32-38%.  As of 2015, reticulosarcoma, also known as lymphoma of the bone, has 89% five-year survival in stage I if treated with chemo + radiotherapy. But bone lymphoma is also often a sign of metastatic disease, which would have lower survival rates. 16 of the 47 cases in the 1971 study were metastatic.  The failures of Coley toxins tended to be briefer courses rather than longer ones, and intramuscular rather than intratumoral or intravenous.

More recent attempts to replicate the effectiveness of Coley toxins (also known as mixed bacterial vaccine or MBV) have been less successful. A Chinese study[20] in 1991 of 86 patients with hepatocellular carcinoma were randomized to either MBV or control; there was a nonsignificant trend towards better survival in the MBV group. An uncontrolled study [21] of MBV in 12 patients with refractory malignancies found one partial response, and a “dramatic improvement in performance status and disease stabilization” in a patient with AIDS and Kaposi’s sarcoma.  An uncontrolled German study[22] of 12 patients with NY-ESO-1 expressing tumors (6 melanoma, 2 sarcoma, 2 prostate cancer, 1 with head and neck cancer, 1 with bladder cancer) found one partial response after treatment with MBV.

Coley toxins are popular in alt-med circles, and the effects haven’t been replicated in a modern experiment. However, the historical data does seem to show dramatic effects, and it is mechanistically plausible that bacterial vaccines stimulate the immune system.  A point in defense of Coley toxins is that the few existing modern trials don’t copy what Coley did (repeated, intratumoral injections, calibrated to produce fever, with patients selected for sarcoma).  A controlled trial of Coley toxins under those conditions could be extremely valuable.

Endotoxin and Tumor Necrosis Factor

Endotoxin, also known as lipopolysaccharide, is a molecule found on the membrane of Gram-negative bacteria, which elicits a strong immune response in the host.  Endotoxin causes the release of inflammatory cytokines, fever, and in extreme cases septic shock.  It also has antitumor effects, strengthening the case that bacterial infection can cause cancer regression.

A study [24] of intravenously administered endotoxin found that it caused partial responses in 2 out of 18 patients with colorectal cancer.  Intravenous endotoxin caused one complete remission out of 37 patients [25] with colorectal and non-small-cell lung cancer.

Parenteral injection of bacterial endotoxin in mice reliably [26] causes necrosis of experimental tumors.  Spindle cell sarcoma (SA-1) and fibrosarcoma (Meth A) regress completely, while CaD2 (breast carcinoma) and BP3 (another kind of fibrosarcoma) grow unchanged.  Necrosis occurred in all tumors. Endotoxin does not cause tumor regression in T-cell deficient mice.  In mice whose tumors regressed, subsequent injection with cells from the same tumor did not grow, suggesting that sensitized T-cells are the cause of the tumor resistance.

Environmental exposure to endotoxin in Chinese workers in the textile industry is inversely associated with lung cancer risk (HR = 0.60, p = 0.002)[27].

The anti-tumor activity of endotoxin appears to be associated with tumor necrosis factor, or TNF.  In 1975 [28] it was observed that Bacillus Calmette-Guerin-infected mice treated with endotoxin produced a substance, dubbed TNF, which was as effective as endotoxin itself in causing regression of transplanted sarcomas in mice.

TNF is too dangerous to give patients intravenously, but it is effective on sarcomas when administered locally when combined with cytotoxic chemotherapy.  High-dose TNF-alpha administered[29] in isolation perfusion of the limbs of 23 patients with metastatic melanoma or soft-tissue sarcoma found 21 complete responses and 2 partial responses.  In 186 patients [30]with locally advanced soft-tissue sarcomas, isolated limb perfusion with tumor necrosis factor caused complete response in 18%, partial response in 57%, and limb salvage in 82%.

There certainly appears to be a TNF-mediated anti-tumor response, and it is suggestive (and in line with the evidence about spontaneous remissions and Coley toxins) that this effect is associated with fever and especially strong against sarcomas.


I don’t believe there’s a single well-established bacterial cancer treatment right now, but it does seem clear that bacterial infections, particularly febrile infections, and derived substances like endotoxin and TNF, can often cause cancer remissions, especially in sarcomas and some advanced cancers.  Additional “protocol engineering” followed by randomized trials might yield a reliable bacterial therapy.  In the contexts where they are effective, bacterial and bacterial-derived treatments seem to cause meaningful increases in survival time, whereas most targeted chemotherapies do not.


[1]Mellman, Ira, George Coukos, and Glenn Dranoff. “Cancer immunotherapy comes of age.” Nature 480.7378 (2011): 480-489.



[4]Challis, G. B., and H. J. Stam. “The spontaneous regression of cancer: a review of cases from 1900 to 1987.” Acta Oncologica 29.5 (1990): 545-550.

[5]Papac, Rose J. “Spontaneous regression of cancer.” Cancer treatment reviews22.6 (1996): 395-423.

[6]Bodurtha, A. J., et al. “A clinical, histologic, and immunologic study of a case of metastatic malignant melanoma undergoing spontaneous remission.” Cancer37.2 (1976): 735-742.

[7]Ifrah, Norbert, et al. “Spontaneous remission in adult acute leukemia.” Cancer56.5 (1985): 1187-1190.

[8]Wiernik, Peter H. “Spontaneous regression of hematologic cancers.” National Cancer Institute Monograph 44 (1976): 35-38.

[9]Mitterbauer, M., et al. “Spontaneous remission of acute myeloid leukemia after infection and blood transfusion associated with hypergammaglobulinaemia.”Annals of hematology 73.4 (1996): 189-193.

[10]Hobohm, Uwe. “Fever and cancer in perspective.” Cancer Immunology, Immunotherapy 50.8 (2001): 391-396.

[11] Maywald, O., et al. “Spontaneous remission in adult acute myeloid leukemia in association with systemic bacterial infection—case report and review of the literature.” Annals of hematology 83.3 (2004): 189-194.

[12]Tocci, G., et al. “Spontaneous remission of hepatocellular carcinoma after massive gastrointestinal haemorrhage.” Bmj 300.6725 (1990): 641-642.

[13]Ruckdeschel, John C., et al. “Postoperative empyema improves survival in lung cancer: documentation and analysis of a natural experiment.” New England Journal of Medicine 287.20 (1972): 1013-1017.

[14] Kleef, Ralf, et al. “Fever, cancer incidence and spontaneous remissions.”Neuroimmunomodulation 9.2 (2001): 55-64.


[16]Wiemann, Bernadette, and Charlie O. Starnes. “Coley’s toxins, tumor necrosis factor and cancer research: a historical perspective.” Pharmacology & therapeutics 64.3 (1994): 529-564.

[17]Richardson, Mary Ann, et al. “Coley toxins immunotherapy: a retrospective review.” Alternative therapies in health and medicine 5.3 (1999): 42.

[18]Miller, Theodore R., and Jesse T. Nicholson. “End results in reticulum cell sarcoma of bone treated by bacterial toxin therapy alone or combined with surgery and/or radiotherapy (47 cases) or with concurrent infection (5 cases).”Cancer 27.3 (1971): 524-548.

[19]Jamshidi, Khodamorad, et al. “Stage IE Primary Bone Lymphoma: Limb Salvage for Local Recurrence.” Archives of bone and joint surgery 3.1 (2015): 39.

[20]Tang, Zhao You, et al. “Preliminary result of mixed bacterial vaccine as adjuvant treatment of hepatocellular carcinoma.” Medical oncology and tumor pharmacotherapy 8.1 (1991): 23-28.

[21]Vas, H. Francis Ha, et al. “Clinical results and immunologic effects of a mixed bacterial vaccine in cancer patients.” Medical oncology and tumor pharmacotherapy 10.4 (1993): 145-158.

[22]Karbach, Julia, et al. “Phase I clinical trial of mixed bacterial vaccine (Coley’s toxins) in patients with NY-ESO-1 expressing cancers: immunological effects and clinical activity.” Clinical Cancer Research 18.19 (2012): 5449-5459.


[24]Engelhardt, Rupert, Andreas Mackensen, and Chris Galanos. “Phase I trial of intravenously administered endotoxin (Salmonella abortus equi) in cancer patients.” Cancer research 51.10 (1991): 2524-2530.

[25]Otto, F., et al. “Phase II trial of intravenous endotoxin in patients with colorectal and non-small cell lung cancer.” European Journal of Cancer 32.10 (1996): 1712-1718.

[26]Berendt, MICHAEL J., ROBERT J. North, and DAVID P. Kirstein. “The immunological basis of endotoxin-induced tumor regression. Requirement for T-cell-mediated immunity.” The Journal of experimental medicine 148.6 (1978): 1550-1559.

[27]Astrakianakis, George, et al. “Lung cancer risk among female textile workers exposed to endotoxin.” Journal of the National Cancer Institute 99.5 (2007): 357-364.

[28]Carswell, E. A., et al. “An endotoxin-induced serum factor that causes necrosis of tumors.” Proceedings of the National Academy of Sciences 72.9 (1975): 3666-3670.

[29]Liénard, Danielle, et al. “High-dose recombinant tumor necrosis factor alpha in combination with interferon gamma and melphalan in isolation perfusion of the limbs for melanoma and sarcoma.” Journal of clinical oncology 10.1 (1992): 52-60.

[30]Eggermont, A. M., et al. “Isolated limb perfusion with tumor necrosis factor and melphalan for limb salvage in 186 patients with locally advanced soft tissue extremity sarcomas. The cumulative multicenter European experience.” Annals of surgery 224.6 (1996): 756.

A note on protocols

Epistemic status: speculative

One thing that stands out about the history of chemotherapy is that what made it broadly effective in leukemia and lymphoma were not drugs but protocols.  Cytotoxic chemo drugs in the 1940’s-60’s had real anti-tumor effects but short-lived remissions until clinicians began administering chemo for longer periods and with combinations of drugs. 

Until working protocols had been set up, chemotherapy was highly controversial.  “As Alfred Gellhorn recently recounted to the authors, the otherwise great clinician Loeb, a giant in the field at the time, had a blind spot when it came to caring for cancer patients and testing chemotherapy. He was fond of saying to Gellhorn, rather openly, “Alfred, you belong to the lunatic fringe.”  This “lunatic fringe” of early chemotherapists persisted in trying different protocols until they got success, despite a heavy death toll.

I’m not sure if someone has made this distinction before, but there seems to be a difference between the “discovery phase” when you observe that some treatment has a desirable property (e.g. a drug has anti-tumor activity) and the “engineering phase” when you figure out how to optimize delivery of that treatment.

In the tech industry, the conventional wisdom is that you need rapid iteration for the “engineering phase” of optimizing the performance of something that already sort of works.

The problem is that rapid iteration on human patients is hard to do, and more so today than in the past.

Rapid iteration is also not particularly suited to the structure of controlled trials.  Trying lots of relatively small changes is harder at large scale and with formal standards of experimental design. It’s more the sort of thing that makes sense for case series.  But it takes a lot of independence on the part of researcher-clinicians, and I suspect that it’s not done enough.

Chemotherapy, Then and Now

Epistemic status: fairly confident

Since I am claiming that cancer research is doing something suboptimal, I’m going to have to examine what progress has actually been made in cancer research, and what results it had.  Here, I’ll focus on the history of chemotherapy.

Early history

“A history of Cancer Chemotherapy” is an excellent article that summarizes the early history.

Chemotherapy is actually a fairly recent development. Until the 60’s, cancer was treated with surgery and radiotherapy. Cure rates plateaued at around 33% due to micrometastases that even aggressive therapy couldn’t reach.

The beginnings of modern chemotherapy were in the Chemical Warfare Service during World War II, which studied chemical weapons and discovered the tumor-regressing effects of nitrogen mustards. The use of nitrogen mustards for treating lymphomas spread rapidly, but remissions were brief.

Sidney Farber, as part of a government drug screening program, and in collaboration with Harriet Kilte, discovered methotrexate and found it effective on children with leukemia.   Also in 1948, George Hitchings and Gertrude Elion discovered 6-thioquanine and 6-mercaptopurine, which would prove useful in the treatment of leukemia.

Post-war, Sloane-Kettering hired almost the entire Chemical Warfare Service for a drug development program. Funding for chemotherapy drug development increased. But the general attitude was still skeptical, since chemotherapy had so far not produced durable remissions.  The first true cure was of the cancer of the placenta, choriocarcinoma, using methotrexate. (The clinical investigator was Min Chiu Li, who was fired for continuing these treatments!)

In the 1950’s, the CCNSC drug development program was also founded; it was the precursor to the modern pharmaceutical industry.  All drugs originated through the CCNSC.  (I was surprised to learn just how centralized the mid-20th-century biomedical research world was.)

Through the 1960’s, chemotherapy was still considered (rightly) to be ineffective. The best results were in leukemia, and about 25% of children with leukemia had remissions, but these were usually measured in months.  The discovery of alkaloids from Vinca rosea and the adoption of combination therapy (multiple drugs taken together) made chemotherapy into a viable strategy.  The “VAMP” program (vincristine, amethopterin, 6-mercaptopurine, and prednisone) got remission rates up to 60% by the end of the decade, and at least half the time these remissions were measured in years.  They were also just starting to mitigate the effects of chemotherapy with platelet transfusions. The late 1960’s also saw the development of the MOMP and MOPP protocols for Hodgkin’s disease (methotrexate/procarbazine, nitrogen mustard, vincristine, prednisone). Complete remission rate went from near 0 to 80% and about 60% of the original patients never relapsed.  Hodgkin’s lymphoma is now regarded as a curable disease.

In the 1970s and beyond there was an expanding role for adjuvant chemotherapy, or chemotherapy in addition to surgery.  Yale Medical School professor and former director of the National Cancer Institute Vincent DeVita  said in an  interview that “At least 50% of the decline in mortality [in colorectal cancer and breast cancer] is due to the application of chemotherapy as an adjunct adjuvant therapy to surgery.”

The first effective chemotherapy regimen for breast cancer was CMF (cyclophosphamide, methotrexate, and 5-fluorouracil), developed in the 1970s.  It was succeeded by AC (doxorubicin + cyclophosphamide).

Adjuvant chemotherapy has meaningful effects on breast cancer.  This 1990 paper says it reduces risk of recurrence by 30% a year, from 4% to 2.8% a year.  This 1989 study says it increases 3-year disease-free survival from 64% to 89%.  The original 1981 Bonnadonna study found it increased 5-year disease-free survival from 45% to 77%.

In the 1970s, the taxanes were developed, the first cytotoxic drugs with efficacy against metastatic breast cancer.  AC + paclitaxel was found more effective than AC alone, but pretty marginally: 70% vs. 65% five-year disease-free survival rate.

And that is more or less where the story of “old-school”, cytotoxic chemotherapy ends. All of these drugs are still in use.  The standard drugs for adjuvant therapy for breast cancer are anthracyclines, taxanes, 5-FU, cyclophosphamide, and carboplatin, all of which were discovered before 1970.

Cytotoxic Chemotherapy

Cytotoxic chemotherapy, the drugs developed in the 1940’s-1970’s, kills cells or inhibits their ability to reproduce; it poisons both cancer and you, but it does more damage to cancer cells than healthy cells because they divide more.

Cytotoxic chemotherapy is currently curative for (some kinds of) lymphoma and leukemia, small cell lung cancer, ovarian cancer, and choriocarcinoma.  It extends survival in many other cancers through adjuvant chemotherapy.  It’s not a cure, but it is genuinely effective.

The following are some examples of cytotoxic chemotherapy drugs:

The nitrogen mustards (discovered 1940’s) alkylate DNA, which makes the cell undergo apoptosis via p53, a protein which scans the genome for defects.  Cisplatin (approved 1978) and cyclophosphamide (approved 1959) work the same way.

Methotrexate (discovered 1947) inhibits dihydrofolate reductase, which is required for catalyzing DNA synthesis (via a few extra steps). Thus, it interferes with cell division.

5-fluorouracil (discovered 1957) blocks the synthesis of thymidine, a nucleoside necessary for DNA replication and hence cell division.

Vincristine (approved 1963) is a mitosis inhibitor.

Doxorubicin (discovered 1960’s) inhibits topoisomerase II, which uncoils DNA in replication.

Paclitaxel (discovered 1967) is a cytoskeletal drug that targets tubulin. Once again, this messes with cell division.

Targeted therapy

Targeted chemotherapy consists of drugs that are meant to specifically kill cancer cells and not healthy cells. Most cancer drugs developed recently are targeted chemotherapies.

The first targeted therapy was approved in 1998; this was imatinib (Gleevec), a tyrosine kinase inhibitor developed by Genentech, for use in myeloid leukemia. It was a huge success, nearly doubling the five-year survival rate from 31% to 59% after it hit the market.  It was also a triumph of “big science”, developed by high-throughput screening.

We are now firmly in the era of targeted therapies, often acting through inhibition of growth factors.  I looked at FDA approvals in oncology from 2000-2015, and found that there were

103 new cancer drugs in the past 15 years

69, or 66%, were targeted

37, or 36%, were growth factor inhibitors.

So, it is indeed true that targeted therapies are the majority of cancer drugs (many of the non-targeted therapies were reformulations or new applications of old cytotoxic chemo drugs) and that growth factor inhibitors take up a large share of the new drugs.

How well do targeted cancer drugs work?

Let’s look at the drugs from 2015, for an example.

Panobinostat is a multiple myeloma drug and a histone deacetylase inhibitor. It is for patients who have received at least two previous treatments.  It is used in combination with other chemotherapy.  It has a partial response rate of 38.5% in one study, 73% in another study (but mostly partial responses.). No apparently available data on survival rates.

Palbociclib is a CDK4 and CDK6 inhibitor for ER-positive and HER2-negative advanced breast cancer.  It increases progression-free survival from 10 months to 20 months but does not increase overall survival times (37.5 months with palbociclib + letrozole vs. 33.3 months with letrozole alone.)

Lenvatinib is a multi-kinase inhibitor approved for iodine-refractory thyroid cancer.  It increases progression-free survival (18.3 months vs. 3.6 months on placebo.)  Only 1.5% had complete responses, however; 63% had partial responses.  It causes no significant difference in overall survival either, whether at study endpoint, or at 6 months, 12 months, or 18 months.

Lonsurf is a combination therapy of trifluridine and tipiracil for metastatic colon cancer.  Trifluridine is an antiviral drug which is also a nucleoside analogue, and tipiracil is a thymidine phosphorylase inhibitor.  Overall survival was 9 months in the treatment group and 6.6 months in the placebo group.

Sonidegib is a Hedgehog signaling pathway inhibitor approved for advanced basal cell carcinoma. It has a 34-36% objective response rate.  I can’t find other information about its efficacy.

Nivolumab is a metastatic non-small-cell lung cancer drug, a PD-1 checkpoint inhibitor. A study found overall survival was 9.2 months with nivolumab vs. 6.0 months with doxetaxel.

Dinutuximab is used for pediatric neuroblastoma.  It looks like so far we only have safety data, not efficacy data.

So, as far as I can tell, not one of these drugs extended life for more than a few months. Part of this is due to the fact that many of these are drugs for refractory or late-stage cancer; this in turn may be due to regulatory or medical-ethics issues that make it hard to directly compare a new drug to the standard-of-care old drugs.

But cytotoxic chemotherapy also had a lot of failures before it found its stride. Perhaps it would be fairer to look at the most common targeted therapies instead of just the most recent ones.


Five-year overall survival rates of 89% in chronic myeloid leukemia patients. Five-year survival rate has doubled over a fifteen-year timespan.  This one definitely works. But it’s special because chronic myeloid leukemia is due to a single chromosomal aberration, the Philadelphia chromosome, which can be targeted precisely with a drug.


This is an HER2 inhibitor for advanced breast cancer.  In combination with docetaxel, vs. docetaxel alone, it had a median overall survival of 31.2 vs 22.7 months, and a 61% overall response rate vs. 34%.  Another study comparing trastuzumab adjuvant chemotherapy vs. chemo alone found 3-year overall survival was not significantly different between the three groups, but progression-free survival was (71% vs. 56%).  The HERA trial, the largest of these so far with 3401 patients in total, found no significant difference in overall 4-year survival, and a statistically significant but small difference in disease-free 4-year survival (78.6% vs. 72.2%.)  A five-year study  found that overall survival with trastuzumab was comparable to HER2-negative patients (5-year survival of about 30%, 1-year survival of 75% and 86%) and higher than HER2-positive patients without trastuzumab (5-year survival rates of about 20%, 1-year survival of 70%).  A study of 4045 women with HER2 positive, non-metastatic breast cancer found a 4-year survival rate with trastuzumab of 93.0% vs. 85.6% without.

Basically, the good studies give modest but real results, and the bad studies say it literally doesn’t extend life.  This looks like ambiguous evidence.


This is a lung cancer drug. In a study of 1215 Asians with advanced pulmonary adenocarcinoma, there was no difference between gefitinib and carboplatin/paclitaxel in overall survival, for either EGFR-positive or EGFR-negative.  228-person study of EGFR-positive lung cancer found no significant difference between gefitinib and carboplatin/paclitaxel.  Study of 1093 patients with advanced non-small-cell lung cancer: no difference in survival times or response rates.


Increases median survival in metastatic colon cancer from 20.0 months to 23.5 months.  And that’s the best-case scenario; median survival differences look worse with different mutation profiles.  Increases median survival in advanced non-small-cell lung cancer from 10 months to 11 months.


In metastatic renal cell carcinoma, extends median overall survival from 21.3 months to 23.3 months.  In non-squamous non-small-cell lung cancer, overall survival was not significantly increased.  In metastatic breast cancer, overall survival was not significantly increased. In metastatic colorectal cancer, improves survival from 14.6 months to 17.9 months.


A meta-analysis shows a statistically significant hazard ratio of 0.63 for chemo + rituximab vs. chemo alone in overall survival for indolent B-cell lymphoma.  Overall survival rate at 3-years from a sample in the post-rituximab era was 75%, vs. 50% from a sample from the pre-rituximab era.  10-year overall survival in elderly patients with diffuse large B-cell lymphoma was 43.5% with R-CHOP vs. 27.6% with CHOP.


For advanced hepatocellular carcinoma, a study of 276 Asian patients found median survival 6.5 months in the treated group vs. 4.2 months in the placebo group.   903 previously treated patients with renal cell carcinoma had comparable overall survival times with sorafenib vs. placebo.


For overall survival in patients with metastatic renal cell carcinoma, 26.4 months with sunitinib vs. 21.8 months with interferon-alpha, in a study of 750 treatment-naive patients.  Another study of metastatic renal cell carcinoma found median survival time 8.7 months for IFN and 17.3 months for sunitinib.


No difference in overall survival between erlotinib + chemo vs. chemo in advanced non-small-cell lung cancer, n = 768.  No difference in overall survival between erlotinib + chemo vs. chemo in EGFR-positive advanced non-small-cell lung cancer.  Study of 1059 patients with advanced non-small-cell lung cancer found no difference in overall survival between erlotinib + chemo vs. chemo.


In metastatic melanoma, study of 676 patients found 10.0 months median overall survival with ipilimumab + gp100, vs 6.4 months with gp100 alone.

Basically, most of these drugs do not extend life more than a few months.  Imatinib is a solitary miracle; rituximab and sunitinib have unspectacular but real effects; trastuzumab is ambiguous; and the other leading targeted chemo drugs appear to be ineffective.

A note on progression-free survival

Increasingly since 1975, progression-free survival has been used as a metric in randomized controlled trials.  The proportion of trials in the Journal of Clinical Oncology using progression-free survival increased from 0% in 1975-1984 to 26% in 2005-2009.  Many drugs show progress in progression-free survival without showing effects in overall survival.   The linked study gives three possible rationales why progression-free survival might improve when overall survival does not:

  • if tumors are small to start with, an increase in tumor size might not have much impact on time to death
  • measurement error: estimates of tumor size are easier to make mistakes on than deaths.  (measurement error, timing error, attrition bias, evaluation bias, etc.)
  • biological explanations: delaying progression may make the tumor more virulent later on, balancing out the overall death rate

So, whenever possible, we want to see improvements in overall survival, not just progression-free survival.

What conclusions can we draw?

My read of the evidence looks fairly similar to James Watson’s — targeted chemotherapy doesn’t look good, with a few exceptions. In other words, this is the leading focus of the pharmaceutical industry with regards to cancer, and its results are distinctly unimpressive.

So far as I can tell, where there has been improvement in cancer mortality since the 1970’s, it is mostly due to people smoking less, early detection of some kinds of cancer, and adjuvant (cytotoxic) chemotherapy, not to targeted chemotherapies.

Why recent developments in chemotherapy aren’t effective is a more speculative matter. There are mechanistic reasons, like Watson’s (growth factors may not be the right strategy). And then there are structural reasons, like DeVita’s.

He notes that it now takes 800 days to get a new cancer protocol approved — at which point your research is out of date. Back in the days before the War on Cancer, when all research was centralized through the NCI, turnaround time was much faster. (Normally we associate decentralization with freedom; but consolidating authority in a small number of people rather than a bureaucratic process can make it easier to get things done quickly and iterate.)  He continues, “But basically, I think what we are trying to say is that ultimately you have to test new things in patients.”

Simple, Upstream, and Decisive: a Heuristic for Medical Progress

Epistemic status: argumentative. This is me laying out an affirmative case; it doesn’t deal with possible objections yet.

Lewis Thomas’s classic 1974 essay, “The Technology of Medicine”, argued that there are three kinds of “technology” in medicine.

  • nontechnology, or “supportive therapy” — basically nursing and reassurance and caring for patients, but not directly affecting the course of disease. Indispensable, expensive, but not really problem-solving.
  • halfway technology — compensating for the damage done by disease when we do not know how to affect its course. Thomas uses organ transplantation as the main example here, as well as all cancer treatment, including surgery, irradiation, and chemotherapy.  It is also very expensive.
  • “decisive technology”, which genuinely reverses or prevents disease. Antibiotics and immunizations are the key examples of this, as are vitamin supplements for vitamin deficiencies, and treatment of endocrine disorders with hormone supplementation. “The point to be made about this kind of technology-the real high technology of medicine–is that it comes as the result of a genuine understanding of disease mechanisms, and when it becomes available, it is relatively inexpensive, and relatively easy to deliver.”

This categorization reminds me of an essay I like about the difference between solving a problem and managing a problem.

“The world of the manager is one of problems and opportunities. Problems are to be managed; one must understand the nature of the problem, amass resources adequate to deal with it, and “work the problem” on an ongoing basis. Opportunities are merely problems that promise to pay off after sufficient work…An engineer believes most problems have solutions. A solution might not be achievable in the short term, but he’s sure somewhere, somehow, inside every problem there lurks a solution. The engineer isn’t interested in building an organisation to cope with the problem. Instead, the engineer studies the problem in the hope of finding its root cause. Once that’s known, a remedy may become apparent which eliminates the need to manage the problem, which no longer exists.”

In Thomas’s classification, nontechnology and halfway technology are “management” approaches to illnesses, while decisive technology is an “engineering” approach.  If you want to actually cure a disease, eradicate it from Planet Earth, you have to have an engineering approach.  Otherwise you’ll be managing it forever.

James Watson (yes, that one, who has spent the last forty years researching cancer), argues a similar case about cancer research in particular.

The decades of the War on Cancer have seen a vast expansion of our understanding of the genetics of cancer; we now know several hundred different genes whose mutations give rise to cancer, and increasing amounts about the molecular pathways that regulate cell growth and differentiation.  We also know that cancer cells develop resistance to most treatments over time as part of this process of accumulating mutations and increasingly abnormal growth patterns.

The most popular approaches to developing cancer drugs tend to involve attacking the growth promoting pathways (such as HER2, RAS, RAF, MEK, ERK, PI3K, AKT and mTOR). But slowing the growth of cancer cells is a priori likely to fail; given that cancer cells adapt, we expect them to develop resistance to such methods. This drives researchers towards increasingly complex cocktails of drugs and “personalized” genomic solutions, in a game of whack-a-mole that can consume arbitrary amounts of resources investigating complex molecular networks, for little practical benefit.  Watson argues that we should focus on strategies that completely kill cancer cells, stop them from dividing, or prevent them from developing resistance — such as Myc inhibitors and p53-activating drugs such as the diabetes drug metformin.

Cancer cells are less genetically stable than normal cells.  This is the principle by which radiation therapy works; it introduces noise into the genome, and because cancer cells are more sensitive to noise than normal cells, the radiation kills the tumor faster than it kills the rest of you. Most traditional chemotherapy also works by introducing “noise”, in the form of reactive oxygen species that disrupt the genome.  Rapidly dividing cells, like cancer cells, suffer more than normal cells (except for normal cells that also divide rapidly, like hair and bone marrow; this is why chemo makes your hair fall out and compromises the immune system.) Of course, these are imperfect solutions, since cancer often becomes resistant to radiotherapy and chemotherapy. Watson’s thesis is that cancer cells protect themselves against reactive oxygen species via antioxidants, and attacking this key ability will make them generically more sensitive to all kinds of cancer treatments; it will essentially make randomness work in your favor against cancer.

Whether or not this theory is valid, the general heuristic seems to be correct. If you want to solve rather than manage cancer, you want an effect simple enough and universal enough that even if there are unknown unknowns (there is always another regulatory gene yet to be discovered) your treatment will still work. You don’t want a treatment strategy that depends on getting every last epicycle right.

However, (to be cynical for a moment) if you wanted to maximize the flow of funding in the long term, you might want to increase complexity without bound…

The classic Hanahan and Weinberg paper, The Hallmarks of Cancer, and its sequel paper, lay out a related model of cancer formation.  The key abnormalities in the development of cancer (which can occur in any order), are

  • self-sufficiency in growth signals
  • insensitivity to anti-growth signals
  • evading apoptosis
  • limitless replicative potential
  • angiogenesis
  • metastasis

This model has a few critical implications for cancer treatment.

The first consequence is that we should not have high confidence that a treatment that attacks only one of these six factors will cause remission.  If the abnormalities can develop in any order, and if acquiring one abnormality makes it easier to acquire others, then blocking one doesn’t necessarily block cancer from becoming metastatic.

The second consequence is that early-stage cancers are easier to kill than late-stage cancers; catching cancer early or preventing it should be more effective than attacking one of the “downstream” consequences of the transition to a cancerous state.  (A major difficulty with this strategy is regulatory: new cancer drugs must usually be tested after conventional chemotherapy fails, so we might never know if an early-stage “chemopreventative” drug would work.)

A third consequence, paradoxically, is that all these changes make cancer cells more vulnerable in certain ways. More reproduction means that their energy demands are higher and their genomes more unstable. Cancer cells use much more glycolysis than normal cells (the famous Warburg effect) and are more vulnerable to genetic damage.  In particular for treating late-stage cancers, it would make sense to focus on the fragility caused by out-of-control growth itself (such as inhibiting glycolysis) rather than trying to slow growth.

In general, we should look for successful treatments (of any disease, including cancer), to be simple, upstream, and decisive.

Simple, because complexity and expense can often be a sign of compensating for continuing damage or omitted factors (like attacking one but not all of the “hallmarks of cancer.”)  Simplicity is also a guard against abuse of multiple hypothesis testing; too many conjunctive details suggest cherry-picking or statistical artifacts.

Upstream, in the sense of focusing on the causal bottleneck, rather than cleaning up after its manifold effects.

Decisive, because effective medical technology works completely.  In other words, look for tumor death, not just shrinkage or slowed growth; look for big effect sizes; look for actual survival improvements, not just changes in some intermediate biomarker or other. Decisiveness is even more important if we consider the impact of statistical malfeasance; it’s much easier for a researcher to massage data into looking just barely significant than to show a clear, dramatic, qualitative result.

Bigness is tempting.  (As a data scientist, I can attest to that.)  Vast quantities of data, infinitely complex models, arbitrary flexibility, all look exciting and powerful.  And there’s the unspoken siren song of overfitting: look, with so many degrees of freedom, you literally can’t be wrong! You can do cool high-tech shit with no risk of embarrassing yourself!  But when it comes to getting things done, as I’ve found even in mundane business settings, this is exactly the wrong mentality; the scientific method is about embarrassing yourself as quickly as possible.

“Simple, upstream, and decisive” is a way of counteracting the temptation towards hiding your failures in bigness.  Of course, it’s a heuristic, not anything like a scientific theory; but viewed through this lens, some research approaches look a priori improbable while others look fruitful.

If some approaches to cancer research are much more likely to work than others, but the scientific community systematically favors the unsuccessful approaches, then the difficulty of curing cancer isn’t just a lack of low-hanging fruit, it’s a strategic problem.  (I haven’t demonstrated yet that this is in fact the case; I hope to provide evidence for that later.)  If it’s a strategic problem and we can distinguish between more-promising and less-promising approaches, then strategic funding might be able to make an impact on cancer even given the vast amount of resources already being poured into it.

Is Cancer Progress Stagnating?

Epistemic Status: a best attempt to give my current understanding.

The War on Cancer began with the National Cancer Act of 1971, and continues to this day. The National Cancer Institute, the largest federally funded organization for cancer research, has a budget of $4.95 billion for 2015; the NIH’s total budget for cancer is $5.39 billion.

But the War on Cancer seems to have disappointing results. Cancer deaths have only fallen by 5% since 1950, at a rate of 200 deaths a year per 100,000 individuals.  (By contrast, heart disease deaths are a third of what they were in 1950, thanks to innovations like statins, stents, and bypass surgery.)

Let’s dig into the cancer numbers to see if this represents real stagnation in medical progress.  It’s possible, for instance, that we’re getting better at treating cancer and it’s just that more people get cancer in the first place, for lifestyle or environmental reasons.

Looking in more depth at the overall cancer numbers from the National Cancer Institute, we see that age-adjusted overall cancer mortality looks a bit better; a 15% decline since 1975, and a 22% decline from the peak of cancer mortality in 1991.  So it looks like cancer deaths were getting worse from the 1950’s through early 1990’s.

Incidence rates also showed a rise from 1975 to a peak in 1991 in males, and a continuing rise in females. Cancer incidence data is sparse before the 1970’s — US population-level data were collected only three times during a period of more than 30 years before 1973.  But cancer incidence was reported to have declined between 1947 and 1970. So the “stagnation” in cancer death rates from 1950 to the present is clearly somewhat confounded by the rise in cancer death rates between 1950 and the early 1990’s.

Now let’s subdivide into some common types of cancer.

Breast cancer has a solid 33% decline in death rates since 1975, beginning in the late 90’s, despite rising incidence.  (It’s one of the most common cancers, at an incidence of 130 per 100,000.)

Prostate cancer, at an incidence of 114 per 100,000, had incidence rise sharply from 1975 to the 90’s, and has been dropping since.  Death rates, likewise, rose and fell over the same timeframe, but are still 35% lower than they were in 1975.

Lung cancer, which has a current incidence of 41 per 100,000, rose dramatically in men to a peak in the early 90’s, and has been falling since; it is less common in women but steadily rising. (This may be a consequence of trends in smoking habits.)  Lung cancer deaths, like overall cancer deaths, rise to a peak in the early 1990’s and decline thereafter; there has been no overall change in lung cancer deaths since the 70’s.

Colon cancer, at 27 per 100,000, has had its incidence drop by 42% since 1985; death rates have dropped accordingly.

Melanoma, at 23 per 100,000, really does have stagnant death rates.

Non-Hodgkins Lymphoma, at 20 per 100,000, has death rates peaking in the late 90’s but no net change between 1975 and the present.

Kidney cancer, at 15 per 100,000, has been rising steadily in incidence since the 1970’s; death rates have been more or less steady. (Some of this may be due to earlier detection of less severe cases.)

Leukemia, at 14 per 100,000, has slightly increasing death rates

Pancreatic cancer, at 13 per 100,000, also has stagnant death rates.

Ovarian cancer, at 12 per 100,000, has declined somewhat in incidence since the 1970’s, and had a roughly comparable decline in mortality.

Cervical cancer, at 6.5 per 100,000, has more than halved in incidence since 1975, and has seen a comparable drop in death rates.

The overall peak and decline in cancer mortality seems to be an artifact primarily of lung cancer incidence, which tracks trends in smoking habits. Breast cancer is the only one of the most common types of cancer that looks like a strong “success story” for treatment — i.e. deaths dropped significantly faster than incidence. Childhood cancer is also a “success story” — deaths dropped in half while incidence slightly increased.  Then there are cases like colon and cervical cancer, where incidence notably declined, perhaps due to the rise in early screening. But on the whole, the figures seem consistent with the hypothesis that the “war on cancer” has not been successful.  In particular, if you look at our ability to treat cancer (as opposed to prevent it, through things like screening or anti-smoking campaigns) the progress looks worse than the raw death numbers would suggest.  (A life saved is a life saved, no matter the means; progress in prevention is a major humanitarian gain.  But it is in some sense less of a technological gain.)

The conventional hypothesis as to why cancer progress has been difficult is that cancer itself is complex and diverse and there is little low-hanging fruit.  “Simplifying principles may not exist”, says an NPR interview with leading cancer researchers. More specifically, the discovery of oncogenes in the early 1970’s led molecular biologists to believe that the onset of cancer was a single “switch” that could be turned off; instead there turned out to be a wide array of oncogenes and tumor suppressor genes, activated in no particular order, and resulting in a great diversity of phenotypes.

This “low-hanging fruit” explanation is the usual one given to explain phenomena like “Eroom’s law”, in which the number of new drugs approved by the FDA per dollar of research spending has steadily declined since 1950.  Biology is complex; progress is hard; that’s it.

The contrarian hypothesis is that cancer research is doing something avoidably wrong.  Usually proponents say something in the vein of “cancer research is too traditionalist” — conservative, slow to innovate, wedded to unsuccessful strategies.  One might assume that this is prima facie absurd — after all, with all the time, money, and human intelligence spent on cancer, if there was a cure out there wouldn’t somebody have found it?  Cancer researchers can’t all be idiots!

In subsequent posts I’ll try to argue that the “cancer isn’t hard” position is at least plausible. This really divides into a negative case — that there are systematic biases in medical research and treatment that push against making progress on cancer — and a positive case — that there are plausible candidates for radical innovation in cancer treatment that deserve more attention.  In an ideal world, this argument would be made by a science journalist or an experienced biologist; I am neither, and so I’ll simply present the facts I’m aware of, with the understanding that it’s pretty incomplete and somebody may come along later and either flesh it out or refute it.