Intended for healthcare professionals

CCBYNC Open access
Research Christmas 2013: Research

The survival time of chocolates on hospital wards: covert observational study

BMJ 2013; 347 doi: (Published 14 December 2013) Cite this as: BMJ 2013;347:f7198

Re: The survival time of chocolates on hospital wards: covert observational study

Recently a study was published in this journal about the lifetimes of donated chocolates in a hospital environment [1]. This study generated much interest. A Google internet search (assessed February 12, 2014) resulted in 986,000 hits using the search phrase “The survival time of chocolates on hospital wards". A critical note however about this study is in order. Because of debatable methods, statistics and results the conclusions of the study are subject to uncertainty.

a) To avoid potential contamination bias, we think the trial should better have been designed as a cluster randomized study [2].

b) The location and display of the two boxes containing the chocolate test products, Roses and Quality Street, in each of the four test areas is critical [3]. Considering the heavy workload of many hospital employees it is likely that the box closest to the entrance door of the test facility will be emptied at the fastest rate. The locations of the boxes at each ward should have been determined using an appropriate randomization scheme.

c) The survival times of the chocolates were measured from the time the boxes were opened. This conflicts with the “intention-to-treat” principle. In our view the survival times should have been measured from the time that the chocolate boxes were placed in the test areas. This appears particularly important because there was a clear difference between the two test brands regarding the mean time until the boxes were opened (19 vs 5 min). This difference was reported to be not significant using the t-test, but absence of a proof of a difference should not be interpreted as a proof of absence of a difference [4].

d) In addition to the previous point, the 95% CI of the mean time until opening of all the 8 boxes used ranged from 0 to 24 minutes. From this it can be calculated that the SD of the opening times equals 14.4 minutes (=12 /2.365 times the square root of 8). This SD is larger than the reported mean value of 12 minutes, which shows that the opening times of the boxes did not follow a normal Gaussian distribution [5]. Therefore the t-test used to compare the mean opening times of the boxes between the brands is not appropriate.

e) There were significant differences between the four investigated wards with regard to the effect size expressing the difference between the two brands, as can be seen from the non-overlapping confidence intervals of the reported separate hazard ratios. No explanation or discussion of these differences is given. Due to the observed heterogeneity of effects the findings are difficult to generalize.

f) The hazard ratio (Roses versus Quality Street) of the two chocolate brands for all wards combined was calculated to be 0.70. This number being less than 1 indicates that the Roses chocolates survived longer than the other brand. But this result contradicts the Kaplan-Meier curves shown which clearly demonstrate that the Roses chocolates survived a shorter time. This raises some questions. Was there perhaps any mislabeling of the test products in the computer database? Or did the data analist by mistake model the event “survival” of the chocolate instead of “death” in the Cox regression survival analysis? Was it verified whether the required proportional hazards assumption in the calculation of hazard ratios was fulfilled?

g) The primary outcome of the study was the median survival time of the chocolates, which was calculated to be 51 minutes. The shown Kaplan-Meier curves however suggest that the median value of the combined groups should be about 100 minutes. Using a well-fitting exponential decay curve the investigators further calculated that the survival half time was 99 minutes. This supports our suspicion that the reported median survival time of 51 minutes is inaccurate.

h) The “loss to follow-up” rate of the chocolates was high (26%). The main reason for this was that observers prematurely ended the continuous observation of the test boxes. This was a clear violation of the protocol which writes (supplementary web information, page 3): “The observers will record data until both chocolate boxes have emptied”. This suggests that the trial was executed in a suboptimal way. Was there any monitoring of the trial by independent qualified persons, especially regarding the conduct, including any chocolate consumption, of the observers?

The authors are complimented for their efforts to get a better understanding of the consumption of donated chocolates at the working place. But, with all respect, in view of some flaws and errors in the trial, we think reliable conclusions are not possible as yet. Further studies are needed.

[1] Gajendragadkar PR, Moualed DJ, Nicolson PLR, Adjei FD, Cakebread HE, Duehmke RM, Martin CA. The survival time of chocolates on hospital wards: covert observational study. BMJ 2013; 347:f7198.

[2] HundleyV, Cheyne H, Bland JM ,Styles M , Barnett CA. So you want to conduct a cluster randomized controlled trial? Journal of Evaluation in Clinical Practice 2010;16:632–638.

[3] Dijksterhuis A, Smith PK, Baaren RB van, Wigboldus DHJ. The unconscious consumer: effects of environment on consumer behavior. Journal of Consumer Psychology 2005;15:193-202.

[4] William Cowper (English poet, 1731-1800), quote: “Absence of proof is not proof of absence”.

[5] Blythe EK, Merhaut DJ. Testing the assumption of normality for pH and electrical conductivity of substrate extract obtained using the pour-through method. HortScience 2007;42:661-669.

Competing interests: No competing interests

13 February 2014
Wim C Hop
Biostatistician, Dept of Biostatistics
Huug W Tilanus, Emeritus professor of Surgery, Dept of Surgery, Erasmus MC, Rotterdam
Erasmus MC, Rotterdam
's-Gravendijkwal 230 3015 CE Rotterdam, The Netherlands