Describing the association between socioeconomic inequalities and cancer survival: methodological guidelines and illustration with population-based data

Figures & data

Table 1 Age-standardized 5-year net survivals by cancer and deprivation quintiles (Q1–Q5, from the less to the more deprived) and EHRs for 1-unit increase of the EDI, in men and women, with their 95% confidence intervals

Cancer	Men						Women
	Q1	Q2	Q3	Q4	Q5	EHR	Q1	Q2	Q3	Q4	Q5	EHR
LOCP	41 [38;43]	38 [36;40]	35 [33;37]	32 [30;33]	29 [27;31]	NL	55 [49;62]	60 [55;64]	58 [54;62]	52 [48;56]	43 [38;47]	NL
Esophagus	13 [11;15]	12 [11;14]	12 [10;14]	12 [11;14]	11 [10;13]	1 [0.98;1.02]	NA	NA	NA	NA	NA	NA
Stomach	27 [24;29]	26 [24;28]	25 [24;27]	25 [23;27]	24 [21;26]	TD	33 [29;37]	32 [29;36]	32 [29;35]	32 [29;35]	32 [28;35]	1 [0.97;1.03]
Colon/rectum	60 [58;62]	58 [57;60]	57 [56;59]	56 [55;57]	53 [51;54]	1.03 [1.01;1.04]	61 [60;63]	60 [59;62]	60 [59;61]	59 [58;61]	58 [56;60]	1.01 [0.99;1.03]
Liver	17 [15;19]	16 [14;17]	16 [15;18]	15 [13;16]	13 [12;15]	1.01 [1;1.03]	NA	NA	NA	NA	NA	NA
Pancreas	8 [6;11]	4 [3;5]	4 [3;5]	5 [4;6]	7 [5;9]	NL and TD	10 [8;13]	9 [7;12]	10 [8;12]	10 [8;11]	9 [7;11]	1.01 [0.99;1.03]
Larynx	53 [48;57]	52 [48;55]	51 [47;54]	50 [47;53]	50 [46;53]	1.01 [0.98;1.04]	NA	NA	NA	NA	NA	NA
Lung	14 [14;15]	13 [13;14]	13 [12;14]	12 [11;13]	10 [10;11]	1.02 [1.01;1.02]	17 [15;19]	17 [15;18]	16 [14;17]	15 [14;17]	14 [12;15]	1.01 [1;1.03]
Melanoma	87 [84;89]	86 [83;89]	83 [81;86]	81 [79;84]	76 [72;80]	1.08 [1.02;1.14]	91 [89;92]	90 [89;92]	90 [88;91]	89 [87;90]	87 [85;89]	1.04 [0.99;1.1]
Breast	NA	NA	NA	NA	NA	NA	87 [86;88]	86 [85;87]	86 [85;87]	85 [84;85]	83 [82;84]	1.03 [1.02;1.05]
Cervix uteri							55 [51;59]	64 [61;66]	66 [63;68]	65 [63;68]	64 [61;67]	NL and TD
Corpus uteri*							73 [70;75]	73 [70;75]	72 [70;74]	73 [71;75]	73 [71;76]	0.99 [0.96;1.02]
Ovary							39 [36;42]	38 [35;40]	37 [35;39]	36 [34;38]	36 [34;39]	1.01 [0.99;1.03]
Prostate	93 [92;93]	92 [91;93]	91 [90;92]	91 [90;91]	88 [87;89]	1.05 [1.02;1.08]
Kidney	69 [66;71]	68 [65;70]	67 [64;69]	65 [63;67]	63 [60;66]	1.02 [0.99;1.05]	63 [60;66]	64 [62;67]	64 [62;66]	64 [61;66]	65 [62;68]	0.99 [0.96;1.03]
Bladder*	63 [60;65]	62 [60;64]	60 [58;62]	58 [56;60]	56 [53;58]	1.03 [1;1.05]	55 [50;59]	52 [47;56]	53 [49;57]	52 [48;56]	49 [44;54]	1.01 [0.98;1.04]
CNS	19 [16;23]	19 [15;22]	19 [16;22]	19 [17;22]	19 [16;23]	1 [0.97;1.03]	22 [19;26]	23 [19;26]	22 [19;25]	22 [19;25]	20 [17;23]	1.01 [0.99;1.04]

Notes: The ICSS weights were used for standardization. The EHR estimates are adjusted on age at diagnosis and year of diagnosis and based on the final model retained (see Materials and methods section) for cancer with a time-constant coefficient and a linear functional form of the EDI. For cancers with an NL functional form and/or a TD coefficient of the EDI, the EHRs are reported in Figures 3 and 4. The bounds of the deprivation quintiles here correspond to those of the deprivation quintiles observed in the general French population.

* Due to sparse data, we merged age groups ([15–45] and [45–55] into a single group [15–55] [adding up the corresponding ICSS weights]) to calculate the age-standardized net survival.

Abbreviations: CNS, central nervous system; EDI, European Deprivation Index; EHR, excess hazard ratio; ICSS, International Cancer Survival Standard; LOCP, lip–oral cavity–pharynx; NA, not available; NL, nonlinear; TD, time-dependent.

Figure 1 Changes over time since diagnosis of the excess mortality hazard for the 10th, 50th, and 90th percentiles of the age distribution (left, middle, and right column, respectively) and for the 10th, 50th, and 90th percentiles of the EDI distribution (curves with black circles, red triangles, and green crosses, respectively) regarding LOCP in men and women, and melanoma in men; patients diagnosed in 2010.

Abbreviations: EDI, European Deprivation Index; LOCP, lip–oral cavity–pharynx.

Figure 2 Changes over time since diagnosis of the excess mortality hazard for the 10th, 50th, and 90th percentiles of the age distribution (left, middle, and right column, respectively) and for the 10th, 50th, and 90th percentiles of the EDI distribution (curves with black circles, red triangles, and green crosses, respectively) regarding stomach and pancreas cancers in men, and cervix uteri; patients diagnosed in 2010.

Abbreviation: EDI, European Deprivation Index.

Figure 3 Excess hazard ratio of the EDI on lip–oral cavity–pharynx cancer in men (A) and women (B) with 95% confidence intervals (shaded area).

Notes: We limited the EDI values on the x-axis to the 5th and 95th percentiles of the observed EDI distribution in the sex-cancer couple. Rug plots indicate the locations of the observed EDI values.
Abbreviation: EDI, European Deprivation Index.

Figure 4 Excess hazard ratio of the EDI at different times after diagnosis for stomach (A) and pancreas (B) cancers in men and cervix uteri (C) cancer in women.

Notes: We limited the EDI values on the x-axis to the 5th and 95th percentiles of the observed EDI distribution in the sex-cancer couple. Rug plots indicate the locations of the observed EDI values.
Abbreviation: EDI, European Deprivation Index.

Table 2 Summary of the guidelines for describing the association between socioeconomic inequalities and cancer survival

Step	Guidelines
Data
	• Use data from a source that provides an unbiased picture of the whole population, such as population-based registries data
	• Use an appropriate ecological deprivation measure, which can be 1) replicated in other countries (for comparison purposes); and 2) based on as small geographical unit as possible
Method
	• Define the excess mortality hazard as your main quantity of interest
	• Use general-population lifetables for the expected mortality hazard and the deprivation-specific ones whenever possible
	• Use flexible parametric multivariable regression models, which enable modeling nonlinear as well as time-dependent log excess hazard ratios for prognostic factors (such as the deprivation index)
	• Take account of the multilevel/hierarchical structure of the data to derive correct statistical inference
	• Use a model-building strategy or an information criterion to eliminate spurious nonlinear and time-dependent log excess hazard ratios
Results
	• Provide model-predicted age-standardized net survivals by deprivation quintile and compare them to the nonparametric estimates (to check the goodness of fit of the model)
	• Give additional and clinically relevant information from the modeling approach: 1) the change with time since diagnosis of the excess mortality hazard for different values of the deprivation index and 2) the excess hazard ratios for the association (eventually nonlinear and/or time-dependent) between the EDI and the excess mortality hazard
	• Quantify the impact of clustering on the excess mortality hazard using the general contextual effect and (whenever possible) an intraclass correlation coefficient

Abbreviation: EDI, European Deprivation Index.