PU580 Unit 3 Case-Control Study Critique Example

Two Hundred Meters Is Not an Exposure: A Unit 3 Study Critique Asking What a Composite Childhood Leukemia Case-Control Study Actually Knew About Each Child, and Why Its Participation Rates Matter More Than Its Odds Ratio

[Student Name]

Purdue University Global

PU580: Environmental Epidemiology

Unit 3 Study Critique

[Instructor Name]

[Date]

The case-control study and its figures are composite; the exposure science, the pooled analyses and the research cited are real.

The Study

Four hundred and twelve incident cases of childhood leukemia from a composite regional registry, 824 population controls sampled from birth records, exposure assigned as residence within 200 meters of a transmission line at birth using addresses and a utility map. Reported odds ratio 1.6, confidence interval 1.1 to 2.3, adjusted for age, sex and area deprivation. Participation: 87 percent of cases, 61 percent of controls.

What Distance Is and Is Not

Distance to a line is a proxy for a time-weighted average magnetic field, and a poor one. Field strength depends on line voltage, current load, which varies hourly and seasonally, conductor configuration, and the geometry of the home, and it falls off rapidly with distance. Two homes 150 meters from the same line can differ severalfold, and a home 300 meters away with a distribution feeder in the wall can exceed both. The pooled analyses that define this literature used measured or calculated fields in microtesla and found an association only above roughly 0.3 to 0.4 microtesla, a level that distance alone cannot identify (Ahlbom et al., 2000).

Problem One: The Exposure Was Never Measured

No field measurements were taken, no load data were obtained from the utility, and no calculation was performed. The study therefore cannot place any child above or below the level at which the pooled literature reports an association, which means its 1.6 cannot be compared with the published estimates at all. A later pooling of newer studies reported a similar pattern and the same dependence on measured or calculated exposure rather than proximity (Kheifets et al., 2010).

Problem Two: Participation, Which Is the Likely Explanation

Eighty-seven percent against 61 percent is the number to look at first. Control participation in these studies is consistently related to socioeconomic position, and socioeconomic position is related both to proximity to transmission corridors and to the likelihood of taking part. A selection analysis of this literature showed that plausible participation differences can generate odds ratios of the size typically reported, with no exposure effect present (Mezei & Kheifets, 2006). The study's odds ratio of 1.6 and its participation gap of 26 points are not independent facts. The second is large enough to produce the first.

Problem Three: Residence at Birth Only

Leukemia is diagnosed at a median of four years in this age group, and families move. Assigning exposure from the birth address ignores every subsequent residence, which both misclassifies exposure and does so differently for families who moved, who differ systematically from those who did not.

Problem Four: Adjustment That Cannot Work

Area deprivation was adjusted for using a tract-level index, which is the only socioeconomic measure available in birth records. Tract-level adjustment leaves substantial individual-level confounding, and in this literature the confounder and the exposure are closely linked, since corridors run through less expensive land. An adjusted estimate here should be read as partially adjusted and the paper should say so.

What a Bias Analysis Would Look Like Here

The critique can do more than assert that participation could explain the finding, and a reviewer should ask for the arithmetic. Suppose exposure prevalence among participating controls is 11 percent, as reported. If non-participating controls were in fact more likely to live near a corridor, at 18 percent, then the true control prevalence across all sampled controls would be about 14 percent, and the odds ratio would fall from 1.6 to roughly 1.2 with an interval crossing one. An 18 percent figure is not a strained assumption: proximity to transmission corridors tracks housing cost, and lower participation among lower-income households is one of the best-documented patterns in survey research. The analysis can be run in either direction, and that is the point. A study with a 26-point participation gap should publish this calculation rather than leaving a reader to do it, and the absence of any quantitative bias analysis is a reporting failure independent of the exposure problem.

What the Study Did Well

Three things, which a fair critique has to record. Case ascertainment through a population registry is complete and avoids referral bias. Controls were sampled from birth records, which is the right frame even though the response rate undermined it. And the authors reported their participation rates, which many studies in this literature did not, and which is what allows this critique to be written.

What Would Make the Study Interpretable

1. Measure or calculate fields at the birth address and at every subsequent address, using utility load data, and report exposure in microtesla against the 0.3 microtesla threshold the literature uses.

2. Characterize non-participants from registry and birth-record variables, and present a bias analysis showing what participation differences could produce.

3. Build residential histories rather than using birth address alone.

4. Report the odds ratio for distance alongside the odds ratio for measured field in the subset with measurements, so a reader can see how the proxy performs.

Why This Literature Keeps Producing Results of This Kind

One paragraph of context helps a reader place the critique. Childhood leukemia is rare, the hypothesized exposure is common and weak, and the plausible effect is small, which is the combination that makes any single study underpowered and vulnerable to small biases. Case-control designs are therefore almost forced, and they bring selection problems with them. Exposure is invisible, cannot be recalled by families, and changes by the hour, so proxies are tempting. And the question carries public concern, which means studies are published and reported whatever their design quality. Those four features together explain why the pooled analyses matter more than any individual paper here, and why a critique should ask where a study would sit in a pooling rather than treating its odds ratio as a standalone finding.

The Judgment

The finding is uninterpretable rather than wrong. A 1.6 with unmeasured exposure, a 26-point participation gap and tract-level adjustment is consistent with a real effect of magnetic fields, with selection alone, and with confounding by socioeconomic position. The published pooled analyses, which did measure exposure, support a weak association above a threshold this study cannot identify, and that is the honest context for the result. A reader should take from this critique not that proximity studies are worthless but that an exposure proxy has to be validated against the quantity the hypothesis concerns before an odds ratio built on it means anything.

References

Ahlbom, A., Day, N., Feychting, M., Roman, E., Skinner, J., Dockerty, J., Linet, M., McBride, M., Michaelis, J., Olsen, J. H., Tynes, T., & Verkasalo, P. K. (2000). A pooled analysis of magnetic fields and childhood leukaemia. British Journal of Cancer, 83(5), 692-698. https://doi.org/10.1054/bjoc.2000.1376

Kheifets, L., Ahlbom, A., Crespi, C. M., Draper, G., Hagihara, J., Lowenthal, R. M., Mezei, G., Oksuzyan, S., Schuz, J., Swanson, J., Tittarelli, A., Vinceti, M., & Wunsch Filho, V. (2010). Pooled analysis of recent studies on magnetic fields and childhood leukaemia. British Journal of Cancer, 103(7), 1128-1135. https://doi.org/10.1038/sj.bjc.6605838

Mezei, G., & Kheifets, L. (2006). Selection bias and its implications for case-control studies: A case study of magnetic field exposure and childhood leukaemia. International Journal of Epidemiology, 35(2), 397-406. https://doi.org/10.1093/ije/dyi245