From the New York Times:

The crudest method to capture a carcinogen’s imprint in a real human population is a large-scale population survey. If a cancer-causing agent increases the incidence of a particular cancer in a population, say tobacco smoking and lung cancer, then the overall incidence of that cancer will rise. That statement sounds simple enough — to find a carcinogen’s shadow, follow the trend in cancer incidence — but there are some fundamental factors that make the task complicated.

The most important of these is life expectancy, which is growing almost everywhere. The average life expectancy of Americans has increased — from 49 in 1900 to 78 in 2011. Several cancers are strongly, often exponentially, age-dependent. An aging population will seem more cancer-afflicted, even if the real cancer incidence has not changed.

But what if we make an “age adjustment” for the population and shrink or expand the cancer incidence to match the changes in age structure? To ask whether cellphones increase the risk of brain cancer, then, we might begin by turning to this question: Has the age-adjusted incidence of brain cancer increased in the recent past?

The quick answer is no. Brain cancer is rare: only about 7 cases are diagnosed per 100,000 men and women in America per year, and a striking increase, following the introduction of a potent carcinogen, should be evident. From 1990 to 2002 — the 12-year period during which cellphone users grew to 135 million from 4 million — the age-adjusted incidence rate for overall brain cancer remained nearly flat. If anything, it decreased slightly, from 7 cases for every 100,000 persons to 6.5 cases (the reasons for the decrease are unknown). In 2010, a larger study updated these results, examining trends between 1992 and 2006. Once again, there was no increase in overall incidence in brain cancer. But if you subdivided the population into groups, an unusual pattern emerged: in females ages 20 to 29 (but not in males) the age-adjusted risk of cancer in the front of the brain grew slightly, from 2.5 cases per 100,000 to 2.6. These cancers appear in the frontal lobe — a knuckle-shaped area immediately behind the forehead and the eye. It is difficult to imagine that cellphones caused these frontal-lobe tumors: how, or why, would a phone’s toxicity have skipped over the area nearest to it and caused a tumor in a distant site? Most epidemiologists and biologists do not find such a tissue-skipping mechanism plausible and most doubt that there is any causal link between frontal tumors and phones.

But a populationwide survey, you might argue, has its limits. The carcinogenic effect of a phone might be so subtle that it never registers in such a survey. A phone may cause cancer after a long lag time — say, 20 years — and it may be too early to look for an effect in a general population. The survey data could be incomplete or of poor quality, thus limiting an epidemiologist’s ability to ever find a discernible link.

Epidemiologists, fortunately, possess a more powerful alternative to uncover a link between a risk factor and cancer. Consider the classic studies that finally revealed the association between tobacco and lung cancer. In the late 1940s, Sir Richard Doll and Sir Austin Bradford Hill, working in London, and Ernst Wynder and Evarts Graham, working in St. Louis, began investigating whether tobacco smoking increased the risk of lung cancer.

Working independently, Doll and Hill, and Wynder and Graham, devised remarkably similar kinds of surveys to reveal a possible link. Using hospital records, they identified a “case” group (a cohort of men with lung cancer) and a matched group of men without lung cancer (a “control” group).

The case group and the control group were asked the same questions, including how much and how often they smoked. By comparing the responses of lung-cancer-afflicted men and nonafflicted men, the two teams of researchers stumbled on a striking association: men with lung cancer had a much longer and deeper history of smoking compared with men without lung cancer.

What if you perform a similar case-control study with cellphones — comparing men and women suffering from brain cancer (cases) and men and women without brain cancer (controls) — looking at their past cellphone use? In 2010, an enormous study, called Interphone, tried to accomplish this task. Setting up the study took years: Interphone recruited participants in 13 countries, ran for a decade and included 5,117 brain-tumor cases and 5,634 controls. The study was coordinated by the World Health Organization and financed primarily by the European Union and cellphone companies, although by agreement industry representatives did not have privileged access to results before publication.

Trials like Interphone are undertaken in the hope that they cleanse the field of doubts. In fact, Interphone achieved just the opposite effect: it ignited even more puzzling questions. Over all, the study found little evidence for an association between brain tumors and cellphones. But when the two cohorts — cancer and no cancer — were subdivided according to the frequency of cellphone use, bizarre results emerged. To start with, there was an apparently decreased risk of brain tumors in regular phone users, compared with rare users or nonusers. In other words, regular cellphone use seemed to reduce the risk of brain tumors. In stark contrast, very high cellphone use (measured as a user’s cumulative call time) seemed to increase the risk of a particular subtype of brain tumor. Needless to say, it is biologically implausible that these results are simultaneously true: how can regular cellphone use protect against cancer while frequent phone use increases risk? To most epidemiologists, including the authors of Interphone, the results point to a systemic flaw in the trial.

Similar case-control studies have examined other kinds of brain tumors, including a rare nonmalignant tumor called an acoustic neuroma. Here, too, the trials have been contradictory. Multiple studies found no association with cellphone use. In contrast, one study from Sweden found an increased risk in people who used their phones for more than 10 years.

How can trials that seem so similar at face value arrive at such disparate and contradictory results? The most likely common problem is bias — built into the very structure of these trials. In a case-control trial, patients are asked to remember their risk of exposure after the fact. In the Interphone study, for instance, participants were asked to recall the extent of their phone use years or even decades in their past. And memory, we now know, is a terribly slippery entity. A patient’s memory of his or her past is a particularly charged and malleable thing; burned into David Reynard’s memory, poignantly, is the shape of the cellphone in his wife’s hand and the imprint of the cancer on her brain.

In fact, our memories turn out to be systematically fragile, especially when we are summoning our past to understand illness. In 1993, a Harvard researcher named Edward Giovannucci set out to measure this phenomenon. Giovannucci identified a cohort of women with breast cancer and an age-matched cohort without cancer, and asked each group about its previous dietary habits. The survey produced a reliable and reasonable trend: women with breast cancer were more likely to have consumed diets high in fat.

But the women in Giovannucci’s study had also completed a dietary survey before their diagnosis of breast cancer. How did a woman’s memory of her diet compare with the actual diet that she recorded before her cancer diagnosis?

Giovannucci’s study illustrates the insidious nature of “recall bias.” In women with no cancer, there was no change between the actual and remembered diet. But women with breast cancer typically recalled a much-higher-fat diet than they actually consumed. The diagnosis of breast cancer had not just changed a woman’s present and the future; it had altered her sense of her past. Women with breast cancer had (unconsciously) decided that a higher-fat diet was a likely predisposition for their disease and (unconsciously) recalled a high-fat diet. It was a pattern poignantly familiar to anyone who knows the history of this stigmatized illness: these women, like thousands of women before them, had searched their own memories for a cause and then summoned that cause into memory.

It is very likely that similar effects undid the Interphone trial: some men and women with brain cancer recalled a disproportionately high use of cellphones, while others recalled disproportionately low exposure. Indeed, 10 men and women with brain tumors (but none of the “controls”) recalled 12 hours or more of use every day — a number that stretches credibility. In a substudy of Interphone, researchers embedded phones with special software to track phone usage. When this log was compared with the “recalled” usage, there were wide and random variations: some users underreported, while others overreported use.

The trouble is that even the largest, longest, best-designed retrospective studies that rely on memory are likely to be riddled by recall bias. Typically, it is not the failure of memory that produces this bias, but its hyperactivity — its desire to explain the uncertainty of the present with the certainty of the past.